Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Sensory, chemical and consumer analysis of
Brettanomyces spoilage in South African wines
Janita J. Botha
Thesis presented in partial fulfilment of the requirements for the degree of
Master of Science in Food Science
Stellenbosch University
Supervisors
M. Muller, Department of Food Science, Stellenbosch University
Dr A. G. J. Tredoux, Institute of Wine Biotechnology, Stellenbosch University
Dr W. J. du Toit, Department of Viticulture and Oenology, Stellenbosch University
Dr A. J. de Villiers, Department of Chemistry and Polymer Science, Stellenbosch University
March 2010
Declaration
By submitting this thesis electronically, I declare that the entirety of the work contained therein is my own, original work, that I am the owner of the copyright thereof (unless to the extent explicitly otherwise stated) and that I have not previously in its entirety or in part submitted it for obtaining any qualification.
Date: 23/02/2010
Copyright © 2010 Stellenbosch University
All rights reserved
0BSummary
This study focussed on the sensory effects of the main volatile compounds produced by
Brettanomyces yeast causing spoilage in wine. This research firstly aimed to determine the
detection thresholds of eight Brett-related spoilage compounds in wine. The second aim was to
determine the sensory effect of the four most important Brett-related compounds when present
individually in wine. The third aim was to determine the sensory effects of these four compounds
when present in wine in a range of combinations, and to further investigate their effect on
consumer liking. Finally, this project aimed to investigate the incidence of these compounds in a
small range of South African wines.
The sensory detection thresholds of 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, 4-
vinylphenol, 4-vinylguaiacol, isovaleric acid, isobutyric acid and acetic acid were determined.
Apart from 4-ethylcatechol, these values generally agreed well with recent literature where
values determined in wine are available. However, the discrepancies highlighted the importance
of the effect of the medium (wine) when determining sensory detection thresholds. The use of
the median as alternative calculation method was also investigated, and it was found that this
method gives more insightful results than the standard American Society of Testing Materials
(ASTM E679-04) method.
Four compounds, namely 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid
were profiled individually in wine using a trained sensory panel. It was found that all four
compounds caused a suppression of the natural berry-like character in the wine, which induced
a sick-sweet character. 4-ethylphenol contributed Elastoplast™ and leather aromas in the wine,
both of which are commonly associated with Brettanomyces taint. 4-ethylguaiacol added a
medicinal aroma to the wine, and 4-ethylcatechol and isovaleric acid were responsible for
savoury and pungent aromas, respectively.
4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid were also profiled in
combination according to the central composite design. Several univariate and multivariate
methods were applied to the dataset obtained. PARAFAC, a multiway method not widely
utilized regarding sensory data, was applied to the data, the results of which were
complementary to those obtained during univariate and multivariate analyses. It was found that
there is a great deal of interaction between the four compounds profiled in terms of sensory
effects. The most notable was the Elastoplast™ attribute, the intensity of which was affected by
all four compounds. The pungent attribute was also affected by the 4-ethylphenol concentration.
Consumer analysis revealed that some of the samples spiked with Brettanomyces-spoilage
compounds were preferred to the unspiked (control sample). However, no further relationship
could be found between consumer liking and either chemical composition or sensory profile. It is
therefore speculated that consumer liking of Brettanomyces infected wine is driven by more
complex sensory or socio-demographic factors.
Finally, the concentration of 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, 4-vinylphenol, 4-
vinylguaiacol, isovaleric acid, isobutyric acid and acetic acid was determined in a small set of
South African wines, selected to contain a high proportion of wines spoiled by Brettanomyces.
Significant correlations were found between 4-ethylphenol and 4-ethylguaiacol, as well as 4-
ethylphenol and isovaleric acid. However, no correlation could be found between 4-ethylphenol
and 4-ethylcatechol. It is speculated that this lack of relationship is due to the different precursor
profiles present in the analysed wines. This study paved the way for future investigations on the
sensory effects of Brettanomyces spoilage in Pinotage red wine.
1BOpsomming
Hierdie studie het gefokus op die sensoriese invloed van die belangrikste vlugtige komponente
wat deur die Brettanomyces gis geproduseer word en bederf veroorsaak in wyn. Eerstens is
gefokus op die bepaling van die deteksiedrempelwaardes van agt Brett-verwante bederwende
komponente. Die tweede doelwit was om die sensoriese invloed van vier van die mees
belangrike Brett-komponente te bepaal wanneer hulle individueel in wyn voorkom. Die derde
doelwit was om die sensoriese invloed van hierdie vier komponente te bepaal wanneer hulle in
verskillende kombinasies in wyn voorkom, asook die effek daarvan op verbruikervoorkeur.
Laastens is gepoog om die voorkoms van hierdie komponente in ‘n klein seleksie van Suid-
Afrikaanse wyne te bepaal.
Die sensoriese deteksiedrempelwaardes vir 4-etielfenol, 4-etielguaiacol, 4-etielcatechol, 4-
vinielfenol, 4-vinielguaiacol, isovaleraatsuur, isobuteraatsuur en asynsuur is bepaal. Met die
uitsondering van 4-etielcatechol het die waardes oor die algemeen goed ooreengestem met
waardes wat onlangs in die wetenskaplike literatuur gepubliseer is. Die uitsonderings het egter
die belangrikheid van die medium (wyn) gedurende die bepaling van sensoriese
deteksiedrempelwaardes uitgelig. Die gebruik van die mediaan as ‘n alternatiewe
berekeningsmetode is ook ondersoek en daar is gevind dat hierdie metode meer insiggewende
resultate lewer as die standaard American Society of Testing Materials (ASTM E679-04)
metode.
Vier komponente naamlik 4-etielfenol, 4-etielguaiacol, 4-etielcatechol en isovaleraatsuur is
individueel in wyn geprofileer met behulp van ‘n opgeleide sensoriese paneel. Daar is gevind
dat al vier die komponente die natuurlike bessiekarakter in die wyn onderdruk terwyl dit
aanleiding gee tot ‘n onnatuurlike soet karakter. 4-etielfenol is gekenmerk aan Elastoplast™ en
leeragtige aromas in die wyn en beide van hulle word algemeen geassosieer met
Brettanomyces bederf. 4-etielguaiacol het ‘n medisinale aroma tot die wyn toegevoeg en 4-
etielcatechol en isovaleraatsuur het respektiewelik souterige (“savoury”) en sterk (“pungent”)
aromas tot gevolg gehad.
4-etielfenol, 4-etielguaiacol, 4-etielcatechol en isovaleraatsuur is ook in verskeie kombinasies
geprofileer volgens die sentrale saamgestelde ontwerp (“central composite design”). Verskeie
enkelveranderlike en meerveranderlike statistiese analisemetodes is ook op die datastel
uitgevoer. PARAFAC, ‘n meerrigtingsmetode wat nie normaalweg vir sensoriese analise data
gebruik word nie, is ook uitgevoer op die data en die resultate was komplimentêr tot die van die
enkelveranderlike en meerveranderlike analisemetodes. Daar is gevind dat, met betrekking tot
vi
sensoriese effekte, daar noemenswaardige interaksie tussen die vier komponente plaasvind.
Die mees opmerklike hiervan was die Elastoplast™ aroma, waarvan die intensiteit deur al vier
die ander komponente geaffekteer is. Verder is die sterk (“pungent”) aroma beïnvloed deur die
4-etielfenol konsentrasie.
Verbruikersvoorkeur-analise het aangedui dat sommige van die monsters waarby
Brettanomyces bederwende komponente gevoeg is, verkies word bó die kontrole-wyn. Daar
kon egter geen verdere verband gevind word tussen die verbruiker se voorkeur en, nog die
chemise komposisie of sensoriese profiele, van die wyn nie. Daar kan dus gespekuleer word
dat verbruiker voorkeur van Brettanomyces bederfde wyn gedryf word deur meer komplekse en
sosio-demografiese faktore.
Laastens is die konsentrasies van 4-etielfenol, 4-etielguaiacol, 4-etielcatechol, 4-vinielfenol, 4-
vinielguaiacol, isovaleraatsuur, isobuteraatsuur en asynsuur in ‘n seleksie van Suid-Afrikaanse
wyne bepaal. Dié wyne is spesifiek so gekies sodat ‘n aansienlike aantal van hulle met
Brettanomyces bederf was. Betekenisvolle korrelasies is gevind tussen 4-etielfenol and 4-
etielguaiacol, sowel as 4-etielfenol en isovaleraatsuur. Daar is egter geen korrelasie tussen 4-
etielfenol and 4-etielcatechol gevind nie. Daar word vermoed dat hierdie gebrek aan korrelasie
te wyte is aan die voorloperkomponent profiele teenwoordig in die wyne. Hierdie studie het die
weg gebaan vir verdere ondersoeke na die sensoriese effekte van Brettanomyces bederf in
Pinotage rooi wyn.
vii
Acknowledgements
Ms M. Muller for her input and continuous encouragement throughout the project, for always going the extra mile, and inspiring me to the same Dr A. G. J. Tredoux for his input and support, as well as for persevering with the problem of 4-ethylcatechol analysis Dr W. J. du Toit for his input, support and patience. Dr A. J. de Villiers for his input, encouragement and being an exceptional editor. Adriaan Oelofse and Jan Bester for their technical input and general enthusiasm for the project. Frikkie Calitz and Prof van Aarde for their statistical input. Thomas Skov, for his help with PARAFAC and his input into my manuscript. Sensory Panel, without which there would be no thesis. Department of Food Science, SU for use of facilities and hardworking personnel. Especially Shantelle for her vast amount of help in the sensory lab. IWBT & Viticulture and Oenology, SU for use of facilities and personnel. Distell for sponsorship of wine. NRF for bursary. My family, for support, prayers, laughter and repeatedly telling me that I DARE NOT quit! Especially my parents, Johan and Benita, for awakening the interests that led to this thesis, and always encouraging me to pursue excellence. Mareli, for being sister, friend and sometimes parent. My friends (especially Anomien), for listening to me vent, and making me forget why I needed to vent in the first place! My Heavenly Father, for blessing in terms of abilities and opportunities. I am constantly in awe of the Master plan.
viii
Notes
The language and style used in this thesis are in accordance with the requirements of the
scientific journal, International Journal of Food Science and Technology.
This thesis represents a compilation of manuscripts where each chapter is an individual entity
and therefore some repetition between chapters may occur.
ix
Table of contents
Declaration.................................................................................................. ii
Summary .................................................................................................... iii
Opsomming ................................................................................................ v
Acknowledgements.................................................................................. vii
Notes ........................................................................................................ viii
Chapter 1: Introduction............................................................................ 1
1 BRETTANOMYCES: THE CURRENT SITUATION ......................................................... 1
2 RESEARCH AIMS ............................................................................................................ 3
3 REFERENCES.................................................................................................................. 4
Chapter 2: Literature Review: Brettanomyces in red wine................... 7
1 INTRODUCTION............................................................................................................... 8
2 INCIDENCE OF BRETTANOMYCES SPOILAGE ........................................................... 9
3 MICROBIOLOGICAL FACTORS CONTRIBUTING TO BRETT SPOILAGE ................ 10
3.1 Brettanomyces contamination .................................................................................. 10
3.2 Factors influencing Brettanomyces spp. growth .................................................... 10
4 SENSORY CHARACTERISTICS OF BRETTANOMYCES SPOILAGE ........................ 11
5 CHEMICAL COMPOUNDS RESPONSIBLE FOR BRETT CHARACTER .................... 12
5.1 Volatile phenols and sensory impact ....................................................................... 12
5.2 Volatile phenol breakdown products........................................................................ 15
x
5.3 Other compounds potentially associated with Brett character ............................. 16
6 FACTORS INFLUENCING LEVELS OF VOLATILE PHENOLS IN WINE .................... 19
6.1 Factors influencing volatile phenol synthesis......................................................... 19
6.2 Volatile phenol sorption............................................................................................. 19
7 SENSORY AND CHEMICAL ANALYSIS METHODOLOGIES...................................... 20
7.1 Sensory methodologies associated with analysing Brett character..................... 20
7.1.1 Analytical tests ....................................................................................................... 21
7.1.2 Hedonic tests .......................................................................................................... 24
7.1.3 Testing for association using statistical analyses .............................................. 26
7.2 Chemical methodologies for analysing compounds associated with Brett character .................................................................................................................................. 27
8 SUMMARY...................................................................................................................... 30
9 REFERENCES................................................................................................................ 30
Chapter 3: The determination of detection thresholds of eight Brettanomyces-related compounds in Pinotage red wine using two calculation methods................................................................................. 39
1 INTRODUCTION............................................................................................................. 40
2 MATERIALS AND METHODS........................................................................................ 43
2.1 Samples....................................................................................................................... 43
2.2 Determination of detection threshold levels............................................................ 44
2.2.1 Subjects and training ............................................................................................. 45
2.2.2 Determination of detection thresholds................................................................. 46
2.2.3 Analysis of data ...................................................................................................... 47
3 RESULTS AND DISCUSSION ....................................................................................... 49
3.1 Comparison between different calculation methods .............................................. 49
3.2 Comparison to literature............................................................................................ 55
4 CONCLUSIONS.............................................................................................................. 57
5 REFERENCES................................................................................................................ 58
xi
Chapter 4: Sensory profiling of four separate Brett-related compounds in Pinotage red wine: 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid ........................................................... 62
1 INTRODUCTION............................................................................................................. 63
2 MATERIALS AND METHODS........................................................................................ 64
2.1 Wine samples.............................................................................................................. 64
2.2 Chemicals and spiking............................................................................................... 64
2.3 Singular profiling of samples .................................................................................... 66
2.4 Data analysis............................................................................................................... 68
3 RESULTS AND DISCUSSION ....................................................................................... 68
3.1 4-ethylphenol .............................................................................................................. 68
3.2 4-ethylguaiacol ........................................................................................................... 72
3.3 4-ethylcatechol ........................................................................................................... 75
3.4 Isovaleric acid............................................................................................................. 78
3.5 Overall discussion of common descriptors............................................................. 81
4 CONCLUSIONS.............................................................................................................. 82
5 REFERENCES................................................................................................................ 83
Chapter 5: Investigation into the sensory effects and interactions of four Brett-related compounds in Pinotage red wine............................. 85
1 INTRODUCTION............................................................................................................. 86
2 THEORY OF MULTIWAY METHODS............................................................................ 87
3 MATERIALS AND METHODS........................................................................................ 90
3.1 Central composite design.......................................................................................... 90
3.2 Wine samples.............................................................................................................. 91
3.3 Chemicals and spiking............................................................................................... 91
3.4 Profiling of central composite design combination samples ................................ 93
3.5 Consumer analysis..................................................................................................... 95
3.6 Data analysis............................................................................................................... 96
3.6.1 Central composite design...................................................................................... 96
xii
3.6.2 Consumer panel...................................................................................................... 97
4 RESULTS AND DISCUSSION ....................................................................................... 97
4.1 Profiling of samples ................................................................................................... 97
4.1.1 Berry-like ................................................................................................................. 97
4.1.2 Sick-sweet ............................................................................................................. 100
4.1.3 Elastoplast™ ......................................................................................................... 104
4.1.4 Medicinal ............................................................................................................... 109
4.1.5 Smoky/Savoury..................................................................................................... 111
4.1.6 Pungent ................................................................................................................. 112
4.1.7 Overall effects using different methods of multivariate analysis .................... 114
4.2 Consumer analysis................................................................................................... 124
5 CONCLUSIONS............................................................................................................ 130
6 REFERENCES.............................................................................................................. 132
Chapter 6: Explorative investigation into the incidence of eight Brettanomyces-related spoilage compounds in a selection of South African red wines ................................................................................... 137
1 INTRODUCTION........................................................................................................... 138
2 MATERIALS AND METHODS...................................................................................... 138
2.1 Samples..................................................................................................................... 138
2.2 Chemical analyses ................................................................................................... 138
2.3 Statistical analysis ................................................................................................... 140
3 RESULTS AND DISCUSSION ..................................................................................... 140
3.1 Quantitative results .................................................................................................. 140
3.2 Relationships between compound levels .............................................................. 143
4 CONCLUSIONS............................................................................................................ 146
5 REFERENCES.............................................................................................................. 146
Chapter 7: General discussion and conclusions .............................. 148
REFERENCES.......................................................................................................................... 150
1
Chapter 1: 2BIntroduction
1 BRETTANOMYCES: THE CURRENT SITUATION
Wine can be spoiled by a number of organisms, including lactic acid bacteria, acetic acid
bacteria and yeasts (Du Toit & Pretorius, 2000), as well as tainted from outside sources such as
cork (Prescott et al., 2005). Wine spoilage generally results in a decrease in the quality of wine,
and, if it is not detected before distribution, disappointment on the part of the wine consumer.
However, consumer disappointment is directly related to the sensory effect of these wine taints,
and not necessarily to the levels of spoilage compounds or spoilage organisms found in a wine.
This leaves the wine industry with a particular problem, as the latter two are relatively easy to
measure, but are not necessarily directly related to the sensory effect – which is the primary
cause of consumer disappointment. Even one instance of disappointment can be enough to
damage the brand of a wine to prevent a consumer from purchasing wines from the same
vineyard, wine region or country of origin. For this reason, it is of utmost importance to not only
define wine spoilage in terms of chemical and microbiological parameters, but also in terms of
sensory and hedonic (consumer enjoyment) parameters (Charters & Pettigrew, 2007).
Brettanomyces is a wine spoilage organism that is related to several wine faults, most
notably the fault originally known as phenolic off-flavour (Chatonnet et al., 1992) or Brett
characterF
1F. This flavour can be described as horsey, leathery, medicinal, band-aid™, smoky or
savoury (Chatonnet et al., 1992; Licker et al., 1999; Wirz et al., 2004; Norris, 2004; Saurez et
al., 2007; Romano et al., 2009). Although Brettanomyces is considered a spoilage organism
and causes an objectionable flavour in red wine when its spoilage compounds are present in
high levels, low levels of Brett character is sometimes considered to add complexity to a wine.
While the physiological characteristics of the yeast are generally well explored, there has
recently been a renewed interest in it, especially in terms of molecular detection methods for the
yeast (Campolongo et al., 2009; Oelofse et al., 2009), chemical detection methods for its
spoilage compounds (Boutou & Chatonnet, 2007; Carillo & Tena 2007; Cyncar et al., 2007;
Fariña et al., 2007; Larcher et al., 2007; Pizarro et al., 2007; Larcher et al., 2008; Hisomoto et
al., 2009) and its sensory effects (Curtin et al., 2008; Cliff & King, 2009; Romano et al., 2009).
This can partially be ascribed to the fact that more sophisticated methodologies have been
developed, which have made these studies possible. Furthermore, there has been an increased
awareness of sensory science, the appropriate methodologies for performing sensory tests and
the possibilities of what can be achieved with sensory science (Tuorila & Monteleone, 2009).
1 The terms Brett character, “Brettyness” or simply “Brett” are used in literature to refer to the sensory effect of wine spoiled by Brettanomyces. As far as possible, the term Brett character will be used throughout.
2
The two main spoilage compounds traditionally associated with Brettanomyces are 4-
ethylphenol and 4-ethylguaiacol. The aroma of 4-ethylphenol is associated with leather/band-
aid™ while 4-ethylguaiacol has a medicinal/spicy smell associated with it (Saurez et al., 2007).
These compounds are generally considered indicator compounds for spoilage by
Brettanomyces. The combined rejection threshold of a total combined concentration of these
two compounds of 426 µg/L is generally used as diagnostic criterion for wine potentially spoiled
with Brettanomyces (Chatonnet et al., 1992). However, poor qualitative correlations have been
found between the presence of these compounds and their sensory effects (Romano et al.,
2009). This may be ascribed to sensory interactions between these compounds, some of the
other compounds associated with Brettanomyces spoilage, as well as other compounds
originating from the grapes, alcoholic fermentation and ageing.
Two other compounds of particular interest are isovaleric acid and 4-ethylcatechol.
Isovaleric acid is formed by the metabolism of L-leucine (Harwood & Canale-Parola, 1981), and
there has been much debate on its sensory effect on wines spoiled by Brettanomyces (Licker et
al., 1999; Fugelsang & Zoecklein, 2003; Romano et al., 2008; Romano et al., 2009). 4-
ethylcatechol is formed in an analogous manner to 4-ethylphenol and 4-ethylguaiacol, but has
only recently been linked to Brettanomyces spoilage due to the fact that 4-ethylcatechol cannot
be detected by the same chemical analysis methods as 4-ethylphenol and 4-ethylguaiacol due
to its lower volatility (Hesford & Schneider, 2004; Hesford et al., 2004; Carillo & Tena, 2007).
The sensory effects of 4-ethylcatechol in wine are still poorly understood (Curtin et al., 2008;
Larcher et al., 2008). In the South African wine industry, chemical diagnosis of Brettanomyces
spoilage is generally limited to testing for elevated 4-ethylphenol and 4-ethylguaiacol
concentrations. However, Pinotage a uniquely South African wine cultivar, contains significantly
higher levels of the precursors of 4-ethylcatechol (De Viliers et al., 2005). This makes
investigation into the sensory and chemical effects of this yeast in Pinotage wine of utmost
importance
Limited studies have been performed on the effect of Brettanomyces-related spoilage
compounds and the acceptability of wines (Etiévant et al., 1989; Chatonnet et al., 1992),
although recent studies by the Australian Wine Research Institute (AWRI) have found that
Australian consumers find wines tainted with Brettanomyces less acceptable than untainted
wines (Lattey et al., 2007; Curtin et al., 2008).
The field of sensometrics investigates relationships between chemical profiles, sensory
descriptors and hedonics. When hedonics is mapped against other wine characteristics, the
statistical technique is known as Preference Mapping. This technique has been applied to wines
(Frøst & Noble, 2002), but is not generally used to investigate wine taints. However, the fact that
this technique can use a bipolar scale such as the nine-point hedonic scale, and can therefore
be used to measure both positive and negative hedonic responses, makes this technique ideal
3
for investigating the effect of Brettanomyces spoilage compounds and their effects on the
consumer preference of wines.
The stigma that is attached to Brettanomyces by the wine industry is a hurdle faced by
researchers, as winemakers are embarrassed to admit that they might have a problem
regarding Brett and therefore reluctant to co-operate with research. In spite of the widespread
denial by winemakers that they have a Brett problem, Australian winemaker Brian Crosser
commented that Brett character was prevalent amongst wines tasted at a recent prestigious
South African wine show (Eedes, 2009). The anecdotal prevalence of this defect makes
investigation into Brettanomyces spoilage in the South African context relevant and absolutely
necessary.
.
2 RESEARCH AIMS
The overall aim of this study was to systematically investigate the sensory effects of
Brettanomyces spoilage compounds in South African wine. This was done by combining
sensory profiling with chemical and consumer analysis. The specific aims of the project can be
summarised as follows:
i) To determine the sensory detection thresholds of eight compounds originating from
Brettanomyces infection in Pinotage wine (Chapter 3).
ii) To determine the sensory effects of 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and
isovaleric acid on the sensory profile of Pinotage red wine spiked with these compounds
(Chapter 4).
iii) To determine the sensory interactions between the above-mentioned four compounds
on the sensory profile of Pinotage red wine spiked with these compounds (Chapter 5).
iv) To determine the effect of the above-mentioned four compounds on consumer
acceptance of Pinotage red wine spiked with these compounds (Chapter 5).
v) To investigate the prevalence of Brettanomyces-related spoilage compounds in South
African red wines (Chapter 6).
4
3 REFERENCES
Boutou, S. & Chatonnet, P. (2007) Rapid headspace solid-phase microextraction/gas
chromatographic/mass spectrometric assay for the quantitative determination of some of
the main odorants causing off-flavours in wine. Journal of Chromatography A. 1141, 1 -
9 .
Campolongo, S., Rantsiou, K., Giordano, M., Gerbi, V. & Cocolin, L. (2009) Brettanomyces
bruxellensis incidence and diversity in Italian wines as determined by molecular methods
(abstract). American Journal of Enology and Viticulture. 60, 398.
Carrillo, J. D. & Tena, M. T. (2007) Determination of ethylphenols in wine by in situ derivitisation
and headspace solid-phase microextraction-gas chromatography-mass spectrometry.
Analytical and Bioanalytical Chemistry. 387, 2547 - 2558.
Charters, S. & Pettigrew, S. (2007) The dimensions of wine quality. Food Quality and
Preference. 15, 997 - 1009.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
Cliff, M. A. & King, M. C. (2009) Influence of serving temperature and wine type on perception of
ethyl acetate and 4-ethyl phenol in wine. Journal of Wine Research. 20, 45 - 52.
Curtin, C. Bramley, B. Cowey, G. Holdstock, M. Kennedy, E. Lattey, K. Coulter, A. Henschke, P.
Francis, L. Godden, P. Sensory perceptions of 'Brett' and relationship to consumer
preference. Blair, R.J.; Williams, P.J.; Pretorius, I.S. Proceedings of the thirteenth
Australian wine industry technical conference, 29 July-2 August 2007, Adelaide, SA. :
207-211; 2008.
Cyncar, W., Cozzolino, D., Dambergs, B., Janik, L. & Gishen, M. (2007) Feasibility study on the
use of a head space mass spectrometry electronic nose (MS e_nose) to monitor red
wine spoilage induced by Brettanomyces yeast. Sensors and Actuators. 124, 167 - 171.
De Villiers, A., Majek, P., Lynen, F., Crouch, A., Lauer, H. & Sandra, P. (2005) Classification of
South African red and white wines according to grape variety based on the non-coloured
phenolic content. European Food Research and Technology. 221, 520 - 528.
Du Toit, M. & Pretorius, I. S. (2000) Microbial spoilage and preservation of wine: Using weapons
from nature's own arsenal - a review. South African Journal of Enology and Viticulture.
21, 74 -96.
Eedes, C (2009). Brett in evidence at Old Mutual Trophy Wine Show 2009 [WWW document].
HUhttp://www.winemag.co.za/article/hallo-darkness-my-old-friend-2009-06-24 UH 29 October
2009.
Etiévant, P. X., Issanchou, S. N., Marie, S., Ducruet, V. & Flanzy, C. (1989) Sensory impact of
volatile phenols on red wine aroma: influence of carbonic maceration time and storage.
Sciences des aliments. 9, 19 - 31.
5
Fariña, L., Boido, E., Carrau, F. & Dellacassa, E. (2007) Determination of volatile phenols in red
wines by dispersive liquid-liquid microextraction and gas chromatography-mass
spectrometry detection. Journal of Chromatography A. 1157, 46 - 50.
Frøst, M. B. & Noble, A. C. (2002) Preliminary study on the effect of knowledge and sensory
expertise on liking for red wines. American Journal of Enology and Viticulture. 53, 275 -
284.
Fugelsang, K. C. & Zoecklein, B. W. (2003) Population dynamics and effects of Brettanomyces
bruxellensis strains on Pinot noir (Vitis vinifera L.) wines. American Journal of Enology
and Viticulture. 54, 294 - 300.
Harwood C. S. & Canale-Parola, E. (1981) Adenosine-5'-Triphosphate-yielding pathways of
branched-chain amino acid fermentation by a marine Spirochete. Journal of
Bacteriology. 148, 117 - 123.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11-13.
Hesford, F., Schneider, K., Porret, N. & Gafner, J. (2004) Identification and analysis of 4-ethyl
catechol in wines tainted by Brettanomyces off-flavor (abstract). American Journal of
Enology and Viticulture. 55, 304A.
Hisomoto, M., Okuda, T., Nishimoto, S., Tani, K., Tachibana, M., Koizumi, H., Kiba, N. &
Yokotsuka, K. (2009) Determination of 4-vinylcatechol in wine by HPLC-DAD coupled
with flourescence detection (abstract). American Journal of Enology and Viticulture. 60,
402.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determinination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Larcher, R., Nicolini, G., Puecher, C., Bertoldi, D., Moser, S. & Favaro, G. (2007) Determination
of volatile phenols in wine using high-performance liquid chromatography with a
coulometric array detector. Analytica Chimica Acta. 582, 55 -60.
Lattey, K. A., Bramley, B. R., Francis, I. L., Herderich, M. J. & Pretorius, S. (2007) Wine quality
and consumer preferences: understanding consumer needs. Australian and New
Zealand Wine industry Journal. 22, 31 - 39.
Licker, J. L, Acree, T. E. & Henick-Kling, T. (1999) What is "Brett" (Brettanomyces) flavour? A
preliminary investigation. In: Chemistry of Wine Flavour. ACS Symposium Series (edited
by A. L. Waterhouse & S. E. Ebeler), Pp 96 - 115. Washington DC: American Chemical
Society.
Norris, L. (2004) Unraveling the mystery of Brettanomyces flavor (abstract). American Journal of
Enology and Viticulture. 55, 304A.
6
Oelofse, A., Lonvaud-Funel, A. & du Toit, M. (2009) Molecular identification of Brettanomyces
bruxellensis strains isolated from red wines and volatile phenol production. Food
Microbiology. 26, 377 - 385.
Pizarro, C., Pérez-del-Notario, N. & González-Sáiz, J. M. (2007) Determination of Brett
character responsible compounds in wines by using multiple headspace solid-phase
microextraction. Journal of Chromatography A. 1143, 176 - 181.
Prescott, J., Norris, L., Kunst, M. & Kim, S. (2005) Estimating a "consumer rejection threshold"
for cork taint in white wine. Food Quality and Preference. 16, 345 - 349.
Romano, A., Perello, M. C., de Revel, G. & Lonvaud-Funel, A. (2008) Growth and volatlie
compound production by Brettanomyces/Dekkera bruxellensis in red wine. Journal of
Applied Microbiology. 104, 1577 - 1585.
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of “Brett character”. Food Chemistry. 114, 15 - 19.
Saurez, R., Saurez-Lepe, J. A., Morata, A. & Calderon, F. (2007) The production of
ethylphenols in wine by yeasts of the genera Brettanomyces and Dekkera: A review.
Food Chemistry. 102, 10-21.
Tuorila, H. & Monteleone, E. (2009) Sensory food science in the changing society:
Opportunities, needs, and challenges. Trends in Food Science & Technology. 20, 64 -
62.
Wirz, D. O., Heymann, H. & Bisson, L. F. (2004) Descriptive analysis of Brettanomyces-infected
Cabernet Sauvignon wines. American Journal of Enology and Viticulture. 55, 303A.
7
Chapter 2: Literature Review: Brettanomyces in red wine
1 INTRODUCTION............................................................................................................... 8
2 INCIDENCE OF BRETTANOMYCES SPOILAGE ........................................................... 9
3 MICROBIOLOGICAL FACTORS CONTRIBUTING TO BRETT SPOILAGE ................ 10
3.1 Brettanomyces contamination .................................................................................. 10
3.2 Factors influencing Brettanomyces spp. growth .................................................... 10
4 SENSORY CHARACTERISTICS OF BRETTANOMYCES SPOILAGE ........................ 11
5 CHEMICAL COMPOUNDS RESPONSIBLE FOR BRETT CHARACTER .................... 12
5.1 Volatile phenols and sensory impact ....................................................................... 12
5.2 Volatile phenol breakdown products........................................................................ 15
5.3 Other compounds potentially associated with Brett character ............................. 16
6 FACTORS INFLUENCING LEVELS OF VOLATILE PHENOLS IN WINE .................... 19
6.1 Factors influencing volatile phenol synthesis......................................................... 19
6.2 Volatile phenol sorption............................................................................................. 19
7 SENSORY AND CHEMICAL ANALYSIS METHODOLOGIES...................................... 20
7.1 Sensory methodologies associated with analysing Brett character..................... 20
7.1.1 Analytical tests ....................................................................................................... 21
7.1.2 Hedonic tests .......................................................................................................... 24
7.1.3 Testing for association using statistical analyses .............................................. 26
7.2 Chemical methodologies for analysing compounds associated with Brett character .................................................................................................................................. 27
8 SUMMARY...................................................................................................................... 30
9 REFERENCES................................................................................................................ 30
8
1 INTRODUCTION
Wine is aged in order to improve the olfactory, gustatory and visual quality of the product. The
practice of aging wines also allows winemakers to control the release of their products into the
marketplace. Unfortunately, wine spoilage can take place during aging in wooden barrels due to
the presence of undesirable yeasts and other organisms remaining in the pores of wooden
barrels after cleansing and sterilisation. The risk of spoilage is greater when the aging period of
a particular wine is extended. Production trends in the wine industry also have an influence on
this type of spoilage (Suarez et al., 2007).
Current winemaking trends seem to encourage subjecting wine to limited stabilisation
processes. In other words, wines are not subjected to clarification or to physical treatments such
as filtration. The philosophy behind this trend is to maintain high concentrations of aromatic
compounds, pigments and colloids in the wine, in order to allow the product to reach its full
potential. Filtering under certain conditions, for example, has been shown to have a negative
effect on the mouthfeel, body and aroma of a wine (Arriagada-Carrazana et al., 2005). The use
of sulphur dioxide – a preservative that is commonly used in winemaking, is also discouraged.
Another, relatively new trend, is to allow red wine grapes to ripen to a stage of almost over-
ripeness, which increases the polyphenol content of wines, and therefore adds to its character.
This results in a wine with a lower acidity. The higher degree of ripeness (and therefore sugar
content) ultimately leads to a higher alcohol concentration, which causes primary fermentation
and malolactic fermentation to take longer to complete (Kelly, 2003). Unfortunately, the
combined effect of these trends is that the growth of wine spoilage organisms – like
Brettanomyces – is more prevalent than ever before.
Brett character is a wine defect associated with an unpleasant aroma reminiscent of
medicine, leather, horse-sweat or band-aid™ (Elastoplast™) which most often occurs in red
wine (Du Toit et al., 2005). It is caused by spoilage by the yeast Brettanomyces, and its
sporulating form, Dekkera (Loureiro & Malfeito-Ferreira, 2003). The cause of this defect was first
thought to occur during malolactic fermentation, as certain lactic acid bacteria can produce
ethylphenols and the latter are commonly associated with this defect. However, it was found
that these species are not able to produce ethylphenols at the elevated levels associated with
Brett character under typical winemaking conditions (Chatonnet et al., 1995; Suarez et al.,
2007). The fact that the sensory defect associated with Brettanomyces more generally occurs in
red wine, is possibly due to the inability of Brettanomyces to survive in white wine (Barata et al.,
2008), as well as the combined effects of higher pH and increased levels of polyphenols in red
wines. Spoilage of this type has also been found to be more common during the summer
months (Chatonnet et al., 1992). Although the defect is more commonly found in wines aged in
barrels, in particularly old barrels, it has also been found in wines not aged in barrels (Chatonnet
et al., 1992; Rodriguez et al., 2001).
9
The spoilage of wine by Brettanomyces/Dekkera is not a novel occurrence, as the South
African wine industry has been dealing with this problem for the past 40 years. It is also
historically associated with some of the wines made in the Bordeaux and Burgundy regions of
France. A survey done in New Zealand in the early 1970’s showed that Brettanomyces was
widespread in this industry, although it could not be detected in winery interiors or on winery
equipment with the methods then available. No links could be found at that time between the
presence of Brettanomyces and winemaking conditions (Wright & Parle, 1974). However, there
appears to be a recent international rise in incidence of this defect, possibly due to winemaking
trends and modern winemaking techniques which encourage the survival and growth of
Brettanomyces.
2 INCIDENCE OF BRETTANOMYCES SPOILAGE
The wine industry is increasingly concerned about the presence of Brettanomyces in wines
(Rayne & Eggers, 2007), and there have been several anecdotal reports of an increase in
incidence of this defect (Kelly, 2003; Saurez et al., 2007). However, a survey done of Australian
red wines between 1996/1997 and 2002 showed a decrease in mean 4-ethylphenol
concentration in wines from between 1000 and 1200 μg/L in 1996 to approximately 400 μg/L in
2002. This drop in 4-ethylphenol levels is ascribed to the initiative of the Australian Wine
Research Institute (AWRI) to promote optimal SO2 use in wines (Godden & Gishen, 2005).
However, this average level is only slightly lower than the additive threshold, 426 μg/L, the total
concentration at which Chatonnet et al. (1992) determined the combination of 4-ethylphenol and
4-ethylguaiacol to have a negative effect on the sensory quality of a Bordeaux-style red wine.
This means that although the average level has dropped during the above-mentioned period, 4-
ethylphenol concentrations may still be at levels at which their aroma contribution becomes
objectionable in a large number of the wines tested.
It is difficult to accurately determine the “incidence” of Brettanomyces spoilage due to
several factors. The first is the fact that Brettanomyces spoilage can be “measured” or
diagnosed in different ways, and there are often poor correlations between the different
methods. The most obvious way of measuring the incidence of Brettanomyces spoilage is by
determining cell numbers. However, two complications arise here, namely the fact that cell
numbers are not directly related to the levels of ethylphenols found in wines (Fugelsang &
Zoecklein, 2003) and that Brettanomyces populations are difficult to measure accurately using
traditional microbiological methods such as selective plating, due to the viable but non-
culturable nature of this organism (Millet & Lonvuad-Funel, 2000). The levels of ethylphenols in
wine have also been shown to be poorly correlated to the sensory effect that is known as
“Brettyness” or phenolic taint (Romano et al., 2008), which complicates the problem even
further. Finally, Brettanomyces spoilage as such is still a debatable issue, as people differ in
10
their opinion on the perceived contribution of the metabolites produced by this yeast to wine
quality. More recent studies have used combinations of culturing, molecular microbiology (PCR)
and analytical chemistry techniques for the determination of incidence of Brettanomyces
spoilage (Campolongo et al., 2009).
3 MICROBIOLOGICAL FACTORS CONTRIBUTING TO BRETT SPOILAGE
Several factors influence the presence of Brettanomyces in a wine environment. The degree
and method of contamination, as well as nutritional factors and inhibitory agents all play a role.
3.1 9BBrettanomyces contamination
Wooden wine barrels are porous containers that are extremely difficult to clean and even more
difficult to sterilise, and can provide an excellent environment for the survival and transfer of
undesirable microorganisms such as Brettanomyces (Saurez et al., 2007). However, used
barrels are not the only source of contamination, as the sensory defect has been shown to
occur in wines that have had no contact with barrels, as well as wines aged in new barrels
(Chatonnet et al., 1992; Rodriguez et al., 2001). Brettanomyces bruxellensis has been found on
grape skins (Renouf et al., 2007; Renouf et al., 2007a) and this is considered an important
source of contamination. A correlation between the presence of Brettanomyces and Botrytis on
grape skins has also been noted. This may be because excessive heat and moisture both
favour Brettanomyces and Botrytis, rather than a direct interaction between the two species.
Damaged grapes can enhance the development of Brettanomyces on berries, as nutrients
previously trapped in the berries are liberated (Renouf et al., 2007). The reintroduction of
contaminated lees during aging can also introduce Brettanomyces bruxellensis into wine.
Brettanomyces is also commonly found in vats, pumps and equipment that is difficult to sterilise
(Suarez et al., 2007).
3.2 10BFactors influencing Brettanomyces spp. growth
Brettanomyces possesses several competitive advantages over other microbial genera that can
occur in wine. It can survive in the nutritionally poor environment in a wine following completion
of malolactic fermentation, and can even use ethanol as sole carbon source (Uscanga et al.,
2000; Rodriguez et al., 2001; Silva et al., 2004). It is tolerant to slightly lower ethanol levels than
Saccharomyces cerevisea: the upper limit of resistance is said to be at 14.5 - 15 % (Barata et
al., 2008).
11
Brettanomyces has been found to be the only surviving micro-organism in wine after
bottling, due to its ability to survive in the anaerobic conditions (Renouf et al., 2007). However,
Brettanomyces can also grow in aerobic conditions, although its characteristics as an organism
are slightly different than those found under anaerobic conditions.
Under fully aerobic conditions Brettanomyces multiplies more quickly and produces large
volumes of acetic acid and small volumes of ethanol. Semi-aerobic conditions cause the
production of less acetic acid. In aerobic conditions, Brettanomyces has been shown to display
a loss of viability after 200 hours (Ciani & Ferrero, 1997).
The total cell numbers produced during anaerobic conditions are also higher than those
produced during aerobic conditions (Ciani & Ferrero, 1997). Under anaerobic conditions,
Brettanomyces can still ferment even though it is not multiplying. It also produces acetaldehyde
under these conditions, which has the capacity to bind free SO2 in wine, making conditions for
Brettanomyces growth even more favourable (Ciani et al., 2003).
Growth of Brettanomyces bruxellensis can be stimulated by the addition of ammonium
sulphate or yeast extract to a medium (Uscanga et al., 2000). Biotin and thiamine are both
required for growth of this organism (Conterno et al., 2006).
This organism also has the ability to resume growth after an apparent death phase
should the conditions for growth become more favourable, having a viable but non-culturable
state as result (Barata et al., 2008). The “viable” numbers of Brettanomyces can be up to ten
times as large as the culturable population, but these differences are negated as the organism
resumes growth (Millet & Lonvaud-Funel, 2000).
The genus is quite sensitive to SO2, from a level of 0.25 to 0.35 mg/L molecular SO2 (Du
Toit et al., 2005). Barata et al. (2008) reported a slightly lower sensitivity to SO2, and
recommend adjusting the level of molecular SO2 in wine to 1.0 mg/L before barrel ageing. The
presence of oxygen also reduces the sensitivity of Brettanomyces to SO2, but strain differences
also play a role in SO2 sensitivity (Du Toit et al., 2005).
4 SENSORY CHARACTERISTICS OF BRETTANOMYCES SPOILAGE
Sensory descriptors for wine with Brett character include rancid, band-aid, soy, horsey, leather,
tobacco and putrid (Wirz et al., 2004). Brett character also masks inherent fruitiness in wines, as
well as varietal character (Licker et al., 1999; Fugelsang & Zoecklein, 2003; Farina et al., 2007).
The production of acetic acid by Brettanomyces increases the acidity (and therefore sour taste)
of a wine. However, the perception of Brett character is also dependant on wine style and
variety (Saurez et al., 2007; Curtin et al., 2008), and fruity, low-tannin red wines do not tolerate
a large amount of “Brettyness” (Norris, 2004).
12
5 CHEMICAL COMPOUNDS RESPONSIBLE FOR BRETT CHARACTER
Several different compounds have been linked to Brett character, of which the volatile phenols
are most commonly associated with this spoilage defect. Other compounds, most notably
isovaleric acid, have also been linked to this defect by different authors, resulting in some
controversy. These compounds are summarised in XTable 2.1X and XTable 2.2X and will be
discussed in the following sections. In these tables, the term “Brett compound” is used, which
refers to a compound commonly accepted to be related to Brettanomyces spoilage.
Table 2.1. Volatile phenols and breakdown products linked with Brett character. The term “Brett
compound” refers to a compound commonly accepted to be related to Brettanomyces spoilage.
Compound Source Odour Reference Status
4-ethylphenol
(4-EP) Conversion of 4-vinylphenol by vinylphenol reductase
Leather, Elastoplast™ or band-aid™
Chatonnet et a.l, 1992
Main spoilage compound
4-ethylguaiacol
(4-EG) Conversion of 4-vinylguaiacol by vinylphenol reductase
Medicinal Chatonnet et al., 1992
Main spoilage compound
4-ethylcatechol
(4-EC) Conversion of 4-vinylcatechol by vinylphenol reductase
Horsey, smoky
Hesford et al., 2004
Recently accepted
4-vinylphenol
(4-VP)
Conversion of ferulic acid by hydrocinnamate decarboxilase
Almond shell Chatonnet et al., 1992
Accepted as minor spoilage compound
4-vinylguaiacol
(4-VG)
Conversion of p-coumaric acid by hydrocinnamate decarboxilase
Flowery, spicy
Chatonnet et al., 1992
Accepted as minor spoilage compound
4-vinylcatechol
(4-VC)
Conversion of caffeic acid by hydrocinnamate decarboxilase
Phenolic, medicinal, smoky
Hisomoto et al., 2009
Not generally accepted as Brett compound
4-hydroxy-acetophenone
Breakdown product of 4-ethylphenol
Sweet, floral Rayne & Eggers, 2007
Not considered Brett compound
4-acetovanilone Breakdown product of 4-ethylguaiacol
Vanilla Rayne & Eggers, 2007
Not considered Brett compound
5.1 11BVolatile phenols and sensory impact
Limited work has been done on the sensory effects of Brettanomyces spoilage, although some
studies have attempted to link the sensory effects of Brettanomyces with chemical compounds.
The existence of the volatile phenols (4-vinylphenol, 4-vinylguaiacol, 4-ethylphenol and 4-
ethylguaiacol) in red wines has been known since the 1960’s (Etiévant et al., 1989; Chatonnet
et al., 1992), but their presence was originally thought to be of bacterial origin. A threshold at
which 4-ethylphenol would contribute negatively to the wine character has been determined to
13
lie between 1200 and 2400 µg/L, with a level where a phenolic smell became apparent set at
4300 µg/L. The influence of maceration type and aging time on 4-ethylphenol concentration was
also apparent (Eteiévant et al., 1989).
Table 2.2. Compounds that may be associated with Brett character. The term “Brett compound”
refers to a compound commonly accepted to be related to Brettanomyces spoilage.
Compound Source Descriptor Linked by Status
Isovaleric acid Breakdown product of L-Leucine
Sweaty, rancid Licker et al., 1999
Controversial, but generally accepted
Isobutyric acid Breakdown product of L-leucine
Sweaty, rancid Romano et al., 2009
Not considered Brett compound
2-phenyl ethanol N/Ca Spicy Licker et al., 1999
Not considered Brett compound
Ethyl decanoate N/C Plastic Licker et al., 1999
Not considered Brett compound
cis-2-nonenal
trans-2-nonenal N/C Burning tyres
Licker et al., 1999
Not considered Brett compound
Guaiacol N/C Plastic Licker et al., 1999
Not considered Brett compound
4-propylguiacol Extraction from oak wood
Spicy Ferreira et al., 2006
Not considered Brett compound
Acetaldehyde N/C Sherry, nutty, bruised apple
Ciani et al., 2003
Not considered Brett compound
Ethyl acetate N/C Nail polish, fruity Ciani et al., 2003
Not considered Brett compound
a Not confirmed
OH
OO
COOH
OH
HOOC
R
OH
COOH
R
OH
R
OH
R
CO2
Esterase Hydrocinnamate
decarboxilase
Vinylphenol
reductase
4-ethylphenol 4-ethylguiacol 4-ethylcatechol
4-vinylphenol 4-vinylguaiacol4-vinylcatechol
p-coumaric acidferulic acid caffeic acid
R = H: p-coutaric acidR = OCH: fetaric acid R = OH: caftaric acid
Hydroxysterene Ethyl derivativeHydrocinnamic acidCinnamic acid
Figure 2.1. Formation of ethylphenols from cinnamic acid precursors (Oelofse et al., 2008).
14
Some of the first studies linking Brettanomyces bruxellensis to ethylphenols were
performed by Chatonnet et al. (1992, 1993). These studies confirmed that the microbiological
origin of ethylphenols in red wines is indeed the yeast Brettanomyces, and not lactic acid
bacteria present during malolactic fermentation, or the yeast Saccharomyces cerivisae, as
originally thought. Saccharomyces cerivisae (Chatonnet et al., 1993), other yeast species and
Oenococcus oeni can produce 4-vinylphenol and 4-vinylguaiacol from ferulic and para-coumaric
acid, through the action of hydrocinnamate decarboxilase (Chatonnet et al., 1995), but only
Lactobacillus plantarum and Dekkera/Brettanomyces possess the enzyme vinylphenol
reductase which converts the vinylphenols to their respective ethylphenols (Chatonnet et al.,
1995). The action of the two above-mentioned enzymes is shown in XFigure 2.1X. Only
Dekkera/Brettanomyces can produce ethylphenols at the levels found in wines, and have a 50-
60% conversion rate of the available substrate (Chatonnet et al., 1992). The production of
volatile phenols by other organisms is also inhibited by higher levels of polyphenols in wines.
However, their production by Dekkera/Brettanomyces is not inhibited (Chatonnet et al., 1993).
Two theories exist about the reason for the conversion of hydrocinnamic acids to volatile
phenols. The first is that the yeast recovers energy from the decarboxylation/reduction reaction
in the form of an electron gradient, which allows for adenosine-5’-triphospate (ATP) production.
The second is that the yeast detoxifies its environment by this conversion. Phenolic acids have
the capacity to deteriorate the plasmic membrane by destroying the phospholipid bi-layer. The
degradation of phenolic acids may therefore inhibit their action on cell destruction (Renouf et al.,
2007).
Chatonnet et al. (1992) also explored the sensory impact of volatile phenols in red
wines. The perception thresholds of 4-ethylphenol and 4-ethylguaiacol were determined using a
method where the detection threshold was defined as the minimum concentration below which
50% of the large number of tasters (70) failed to detect a difference from the control. The
determination of perception thresholds was done in a hydro-alcoholic model solution, as well as
in water. In the model solution, thresholds were found to be 440 μg/L and 47 μg/L for 4-
ethylphenol and 4-ethylguaiacol, respectively, whereas in water they were found to be 130 μg/L
and 35 μg/L, respectively. They also analysed wines with “phenolic”, “animal” and “stable”
characteristics by Gas Chromatography–Olfactometry (GC-O) in order to determine which
volatile substances were related to the olfactory faults. These authors concluded that 4-
ethylguaiacol, 4-ethylphenol and 4-vinylphenol are associated with the defect. 4-ethylphenol
gave the most intense “stable” characteristic; 4-ethylguaiacol had a spicy/phenolic smell,
whereas 4-vinylphenol had the lowest intensity and had a medicinal smell associated with to it.
The sensory interaction between 4-ethylguaiacol and 4-ethylphenol was also investigated, and
indications of an interaction in terms of detection threshold and sensory impact were observed.
However, this interaction was not clearly defined. It has also been shown that 4-vinylphenol
15
masks fruity nuances below its perception threshold, and that 4-vinylguaiacol adds flowery and
spicy notes to the aroma of a wine (Chatonnet et al., 1992).
Another volatile phenol recently reported in Brettanomyces-infected wines is 4-
ethylcatechol, which has a horsey flavour and is formed from caffeic acid in an analogous
manner to 4-ethylphenol and 4-ethylguaiacol from p-coumaric and p-ferulic acid, respectively
( XFigure 2.1 X). The levels of 4-ethylcatechol produced were related to the grape variety (Hesford
et al., 2004). This finding was confirmed by Larcher et al. (2008). This is most likely due to the
different profiles of hydrocinnamic acid precursors present in the wines produced from different
varieties. It is interesting to note that Pinotage, a uniquely South African cultivar, contains
significantly higher levels of caffeic acid and its precursor, caftaric acid, than other wine cultivars
(de Villiers et al., 2005; Rossouw & Marias, 2004), making this cultivar more susceptible to
spoilage by 4-ethylcatechol. However, Larcher et al. (2008) postulates that 4-ethylcatechol
probably does not have a large negative effect on the sensory profiles of wines. Regardless,
this makes for an interesting topic for investigation.
The detection thresholds determined of 4-ethylcatechol has been determined by three
authors to date, but differ drastically. The three values obtained were: 60 µg/L (Hesford &
Schneider, 2004), 100 – 400 µg/L (Larcher et al., 2008) and 774 µg/L (Curtin et al., 2008). All
these values fall below the maximum level of 1610 µg/L of 4-ethylcatechol that has been found
in wine to date (Larcher et al., 2008), but these discrepancies warrants further investigation into
this compound.
5.2 12BVolatile phenol breakdown products
The stability of ethylphenols in red wine influences their concentrations in wine, as well as the
resulting sensory profile. This aspect is still poorly understood, although it is known that 4-
ethylphenol can be broken down into 4-hydroxyacetophenone via the enzyme 4-ethylphenol
methylenehydroxylase (4EPMH) ( XFigure 2.2 X). Another enzyme, p-cresol methylhydroxylase
(PCMH), breaks down 4-ethylphenol to 4-hydroxyacetophenone and 4-vinylphenol. Similarly,
the breakdown products of 4-ethylguaiacol are 4-vinylguaiacol and 4-acetovanillone ( XFigure
2.3 X). These breakdown products can influence the sensory profiles in Brettanomyces-infected
wines, as they all have their own specific flavours; 4-hydroxyacetophenone has a sweet, floral
aroma, 4-vinylphenol has an almond shell aroma, 4-vinylguaiacol a clove-curry aroma, and 4-
acetovanillone a vanilla aroma (Rayne & Eggers, 2007).
16
OH
R
OH
COOH
R
4EPMH
R = H: 4-ethylphenol R = OCH: 4-ethylguaiacol
4-hydroxyacetophenone 4-acetovanillone
Figure 2.2. Breakdown of ethylphenols by 4EPMH.
OH
R
OH
COOH
R
+
OH
R
PCMH
4-vinylphenol 4-vinylguaiacol
R = H: 4-ethylphenol R = OCH: 4-ethylguaiacol
4-hydroxyacetophenone4-acetovanillone +
Figure 2.3. Breakdown of ethylphenols by PCMH.
5.3 13BOther compounds potentially associated with Brett character
Although volatile phenols are the compounds most commonly associated with Brett character,
they are not the only compounds with sensory implications that have been linked to the
metabolism of Brettanomyces in wines. Licker et al. (1999) reported a study into what they
describe as the “larger picture of Brett character”. Their study investigated three wines obtained
from a winery that were described by the winemaker as “no Brett”, “medium Brett” and “high
Brett”. The wines were shown to contain different levels of 4-ethylphenol in relation to their
described level of Brett character. Sensory profiling of these wines showed that the higher
“Brett” wines had more intense horse sweat, rubber, band-aid™ and plastic aromas, whereas
the “no Brett” wine was dominated by aromas like spicy, earthy, woody, fruity and floral. This
seems to indicate that the effect of Brettanomyces spoilage is not just the addition of some
(seemingly negative) flavours such as band-aid, and horse-sweat, but also the suppression of
other (often positive) wine flavours like fruity and floral.
GC-O coupled with Charm analysis (described by Acree et al., 1984) performed in the
same study revealed that certain compounds other than the volatile phenols appear to add to
the sensory effect that is known as Brett character. These included, in decreasing order of
importance (according to the study) isovaleric acid, an unknown compound, 2-phenyl ethanol,
17
ethyl decanoate, cis-2-nonenal, guaiacol, 4-ethylphenol and trans-2-nonenal. The authors
described their findings as a “snapshot” of the larger picture of Brett character.
The study done by Fugelsang and Zoecklein (2003) contradicts, some and confirms
other findings of Chatonnet et al. (1992) and Licker et al. (1999). In this study a specific wine
was inoculated with different strains of Brettanomyces bruxellensis and the levels of key
metabolites were investigated. They found that although their wines had relatively low levels of
4-ethylphenol and 4-ethylguaiacol, they still had a distinct Brett character, which hints at
synergistic sensory effect between different compounds. This study also indicated no significant
difference in isovaleric acid levels between the control wines and wine inoculated with
Brettanomyces bruxellensis. This suggests that isovaleric acid may not be the most important
odorant associated with Brett character because not all strains of Brettanomyces bruxellensis
produce this compound.
Romano et al. (2008, 2009) shed more light on the role of isovaleric acid in Brett
character. Upon inoculating wines with a strain of Brettanomyces bruxellensis, it was found that
the strain produced significant amounts of compounds other than volatile phenols, including
carboxylic acids (acetic acid, isobutyric acid and isovaleric acid), short chain fatty acids
(hexanoic acid and octanoic acid) and ethyl esters (ethyl octanoate). In a follow-up study
(Romano et al., 2009) no correlation was found between ethylphenol levels in commercial wines
and their degree of sensory “Brettyness”, as at least a third of the wines surveyed that did not
have a noticeable degree of Brett character, but contained ethyl phenols at levels above the
cumulative sensory threshold of 426 μg/L (Chatonnet et al., 1992). A significant difference was
also found between wines inoculated with Brettanomyces bruxellensis and the same wines
spiked with corresponding amounts of 4-vinylphenol, 4-vinylguiacol, 4-ethylphenol and 4-
ethylguaiacol, indicating that other volatile compounds have a detectable impact on wines
spoiled by Brettanomyces bruxellensis. Further chemical investigation revealed a strong
correlation between the production of isobutyric and isovaleric acid and that of ethylphenols,
indicating that these compounds could possibly be considered markers for Brett character.
Isovaleric and isobutyric acids are formed from L-leucine and L-valine respectively, as shown in
XFigure 2.4 X. The detection threshold of ethylphenols in combination with isovaleric and isobutyric
acids were, however, found to be three times higher than that of the ethylphenols by
themselves. The masking sensory effect of these acids may be due to the fact that their “rancid”
and “sweaty” descriptors are quite similar to some of the descriptors usually used for wines with
Brett character. This relationship, which could be either a linguistic confusion or a sensory
interaction, is something which warrants investigation.
This already complex picture is complicated even further by other findings. A study done
by Ferreira et al., (2006) on the kinetics of aroma extraction during aging in oak wood, showed
that certain chemical compounds do not follow “logical” extraction patterns and suggests that
these phenomena happen due to biochemical action on the oak wood. The levels of 4-
18
propylguaiacol extracted from oak wood were found to be positively correlated with higher levels
4-ethylphenol and 4-ethylguaiacol, which are the two compounds most commonly associated
with Brett character. However, this correlation has yet to be confirmed.
The fact that Brettanomyces displays β-glucosidase activity may also impact on wine
flavour. The enzymatic liberation of glycoside hydrolysis products may produce aroma, flavour
and colour changes, which influences wine. The hydrolysis of glycosides may also increase
varietal aroma and flavour in a wine, which would indicate that Brettanomyces has a positive
effect on wine quality. However, the activity of these enzymes in fermentative environments is
limited (Mansfield et al., 2002). Brettanomyces also produces small amounts of acetaldehyde
and ethyl acetate during aerobic fermentation (Ciani et al., 2003). Acetaldehyde has
sherry/nutty/bruised apple descriptors, and ethyl acetate contributes nail polish/fruity aromas to
wine (Swiegers et al., 2005). These compounds further add complexity to sensory character that
may be produced by Brettanomyces spoilage of wine.
L-Leucine
2-KG
Glu
2-Ketoisocaproic acid
CO2
2H
CoA
Isovaleryl CoA
CoA
iP
Isovaleryl-P
ADP
ATP
Isovaleric acid
2-KG
Glu
CO2
2H
CoA
CoA
iP
ADP
ATP
L-Valine
2-Ketoisovaleric acid
Isobutyryl CoA
Isobutyryl-P
Isobutyric acid
Figure 2.4. Production pathway of isovaleric and isobutyric acid from L-leucine and L-Valine
respectively (Harwood & Canale-Parola, 1981).
19
6 FACTORS INFLUENCING LEVELS OF VOLATILE PHENOLS IN WINE
6.1 14BFactors influencing volatile phenol synthesis
It is estimated that a Brettanomyces cell count of approximately 105 cells/mL is required to
trigger the production of ethylphenols in red wines (Fugelsang & Zoecklein, 2003). As previously
mentioned, although higher levels of volatile phenols are produced later in the winemaking
process, when higher numbers of cells are present (Renouf et al., 2007), cell numbers are not
directly related to volatile phenol production (Fugelsang & Zoecklein, 2003). Ethylphenol
production is also influenced by the type of strain present (Fugelsang & Zoecklein, 2003;
Conterno et al., 2006). Ethylphenols are produced in wine during the lag phase of cell growth,
and is inhibited by a lower pH (Romano et al., 2008).
The production of ethylphenols is also influenced by the availability of their
hydrocinnamic acid precursors in the medium, and is proportional to the size of the
Brettanomyces/Dekkera population in the wine (Saurez et al., 2007). Higher amounts of volatile
phenols are produced at lower alcohol concentrations and higher temperatures (Gerbaux et al.,
2002; Saurez et al., 2007). Heating must at the end of maceration also results in higher levels of
volatile phenols, as does the use of extraction enzymes and some clarification enzymes during
winemaking. Many commercial enzymes contain cinnamyl-esterase, which releases
hydrocinnamic acids, leading to higher volatile phenol production (Gerbaux et al., 2002).
Aspergillus moulds present on grape skins, and therefore in the must, also contain these
esterases – which release phenolic acids bound to tartaric acids – and can therefore increase
the free phenolic acid content, leading to higher levels of volatile phenols (Shinohara et al.,
2000).
6.2 15BVolatile phenol sorption
Apart from breakdown to 4-hydroxyacetophenone, 4-acetovanillone, 4-vinylphenol and 4-
vinylguaiacol, levels of 4-ethylphenol and 4-ethylguaiacol in wine can also be decreased by
means of slow partitioning into oak wood and lees (Chassagne et al., 2005; Barrerera-Garcia et
al., 2006; Rayne & Eggers, 2007; Jiménez-Moreno & Ancín-Azpilicueta, 2009). This, along with
the enzymatic breakdown of ethylphenols, results in what is referred to as a “Brett peak”, where
maximum levels of ethylphenols can be found during the summer months, with a decrease
during autumn. Samples taken and subjected to chemical or sensory analysis on either side of
this “Brett peak” may result in a false negative result for Brettanomyces infection.
At typical wine pH values, the hydroxyl groups of 4-ethylphenol and 4-ethylguaiacol are
not dissociated, which reduces the hydrophilic nature of the compounds, making them more
20
likely to associate with organic matter in solution (colloidal or dissolved tannins) or on surfaces,
such as oak barrels (Rayne & Eggers, 2007). Barrera-Garcia et al. (2006) reported that the
sorption of 4-ethylguaiacol and 4-ethylphenol into oak barrels occurs in two phases, with an
initial fast sorption onto the wood surface during the first day, followed by a slower diffusion
between the second and the eighth days. Sorption reached its peak at approximately 1000 mg
of analyte per kg of wood, which is well above the typical amount of ethylphenols found in
wines. The sorption of volatile phenols into yeast lees added to wine has been reported to reach
a maximum at approximately 3 hours of contact. The sorption is accelerated by stirring of the
wine (Chassagne et al., 2005; Jiménez-Moreno & Ancín-Azpilicueta, 2009).
The levels of 4-ethylphenol and 4-ethylguaiacol in a wine can also be reduced through a
method combining reverse osmosis and adsorption (Ugarte et al., 2005). This method uses
tangential-flow filtration equipment, which consists of a membrane filter, a hydrophobic
adsorbent resin, and a reverse-osmosis feed tank. Although this method does not have a
significant effect on colour, anthocyanins or tannins, a significant loss of esters like ethyl
hexanoate and ethyl octanoate has been reported. These compounds are responsible for
apple/banana and pineapple/pear descriptors. However, sensory analysis of wine treated with
this method showed a higher level of fruitiness than an untreated wine. This substantiates the
theory that the volatile phenols suppress fruity character in wines (Ugarte et al., 2005).
7 SENSORY AND CHEMICAL ANALYSIS METHODOLOGIES
7.1 16BSensory methodologies associated with analysing Brett character
There are several factors that complicate sensory studies regarding Brett character in wines.
The first is sample selection. The development of “natural” Brett character usually happens over
a period of several months or years (Saurez et al., 2007), and simulating this in a laboratory
environment is rarely practical. Spiking of wines with chemical compounds known or suspected
to be responsible for Brett character also “confounds” some information, as chemicals may have
sensory interactions that are not well known (Atanasova et al., 2005). Many studies select
samples by means of informal sensory analysis, which usually involves tasting and selecting the
wines that have the highest degree of Brett character. However, this approach has the
subjectivity of the researcher selecting the samples built into it, as the selection depends on
what they consider to be Brett character (Licker et al., 1999; Romano et al., 2009). Another
issue is that methodologies for sensory analysis of wines differ greatly, with very few studies
employing standardised sensory methodologies, which makes results difficult to compare. The
fact that Brett character is associated with barrel aging, and that suppression of inherent
fruitiness is one of its effects, also complicates the sensory analysis. Woody odorants, even
21
when below threshold levels, are known to suppress fruity odorants (Atanasova et al., 2005),
and confusion by judges also occurs between odours of the same family (like woody and
phenolic) (Escudero, 2007), making “true” descriptive analysis even more complicated.
When performing sensory analysis on a wine taint such as Brett character, there are
several strategies that can be followed. These strategies are loosely divided into analytical tests
and hedonic tests, with different standard methodologies existing for both. These tests are
outlined in XFigure 2.5X. Only the tests in this figure relevant to dealing with taints will be
discussed further.
7.1.1 44BAnalytical tests
As can be seen in XFigure 2.5X, sensory analytical tests can be divided into difference tests and
quantitative tests.
Figure 2.5. Outline of different types of sensory tests used for the analysis of taints (Kilcast,
2003).
Sensory testing procedures
Analytical tests Hedonic tests
Preference Difference tests Quantitative tests
Acceptability Paired comparison
Duo-trio
Triangle
Simple descriptive
Profiling
R-index
“Instrumental” measures of
perceived quality
Response of people to
the perceived quality
22
59BDifference tests
The most common application of difference tests in the analysis of wine taints is for the
determination of detection thresholds. Four types of difference tests exist, namely paired
comparison, duo-trio tests, triangle tests and R-index tests.
The paired comparison test poses the question “Is there a difference between the two
samples?” However, this test is not commonly used for the determination of detection
thresholds, as it suffers from response bias (O’Mahony, 1995; Lawless & Heymann, 1998).
Regardless, this test has been used by Hesford and Schneider (2004) for the determination of
the detection threshold of 4-ethylcatechol and a value of 60 µg/L was obtained.
The duo-trio test consists of sets of three samples, of which one is set as a control. This
test poses the question “Which of these samples is different from the control?” This method has
been used to determine the detection threshold of 4-ethylcatechol (Larcher et al., 2008) and has
been used by Fugelsang and Zoecklein (2003) to determine whether wines inoculated with
Brettanomyces were different from a control.
Triangle tests also consist of sets of three samples, but in this test no control is
specified, and the question posed is “Which of these samples is odd?” The three-alternative-
forced choice (3-AFC) test is a special case of the triangle test where the difference between
the samples is known. The 3-AFC has a higher probability of judges identifying the correct
sample (O’Mahony, 1995). The International Organization for Standardisation (ISO) test for
determining detection thresholds, as well as the one prescribed by the American Society for
Testing and Materials (ASTM E-679) (ASTM, 1999) are both based on the 3-AFC test.
A traditional triangle test was used by Ferreira et al. (2000) for the determination of
thresholds for several wine compounds. The ISO 13301 has been used for the determination of
detection thresholds of volatile phenols in wine (Romano et al., 2009), as well as
trichloroanisole (TCA) (Mazzoleni & Maggi, 2007). The ASTM test has been used for
determining the detection threshold of oak lactones (Brown et al., 2006), diacetyl (Martineau et
al., 1995) rotudone (Wood et al., 2008) and 3-isopropyl-2-methoxypyrazine (Pickering et al.,
2007). The widespread use of the ASTM method can be ascribed to the fact that it is a
standardised method which is commonly accepted as correct.
A final type of difference testing is the R-type test. The samples to be tested are
compared to a standard and rated in one of four categories. When performing this type of
difference testing, these categories are “standard”, “perhaps standard”, “perhaps not standard”,
and “not standard”. The results are expressed in terms of R-indices, which represent probability
values of correct discrimination (Lawless & Heymann, 1998; Kilcast, 2003). This test is not
commonly used, but has been applied for the determination of detection thresholds of caffeine
(Robinson et al., 2005) and the determination of off-flavour development time in beef (An et al.,
2009). To date, this method has not been used for the analysis of wine.
23
Regardless of which type of difference testing is involved when detection thresholds are
determined, the medium in which the detection threshold is determined is of utmost importance.
Detection thresholds determined in wine are usually significantly different to those determined in
model solution or water (Le Berre et al., 2007). The wine type and style also influences the
detection threshold (Martineau et al, 1995; Brown et al., 2006). This makes it difficult to compare
detection thresholds reported by different authors.
As previously mentioned, Chatonnet et al. (1992) determined the olfactory perception
threshold of volatile phenols in water, model solution and red wines. These perception
thresholds correspond to the minimum concentration at which 50% of a 70 person jury failed to
taste the difference from a control. A similar test has recently been used for the determination of
volatile phenols in olive oil (Vichi et al., 2009). Although this test seems simple and reliable
enough, the detection threshold determined in such a manner is arbitrary and empirical, as
there is no scientific basis for such a test (Lawless & Heymann, 1998). The perception
thresholds determined by Chatonnet et al. (1992) for 4-ethylphenol and 4-ethylguaiacol in water
were 130 and 25 μg/L, respectively, and in model solution 440 and 47 μg/L, respectively. The
perception threshold of 4-ethylphenol in red wine was determined to be 605 μg/L, while that of
4-ethylguaiacol was 110 μg/L. The arbitrary method of determination is particularly important in
this case as the detection thresholds determined by Chatonnet et al. (1992) are commonly cited
as detection thresholds for these compounds.
60BQuantitative tests
Quantitative tests are used in order to define differences between different products, and can be
used to define the differences between tainted and non-tainted wines. Quantification can occur
either on a category scale or a line scale, which can be either unipolar or bipolar. Category
scales use a defined number of boxes and the scale ends and intermediate points may be given
verbal descriptions. Line scales consist of a line that usually only has verbal anchors at the
ends. Reference points like samples or reference standards may be added to these scales. A
unipolar scale has 0 at one end and the attribute at the other end. On the other hand, bipolar
scales have opposite attributes at different ends (Lawless & Heymann, 1998; Kilcast, 2003).
However, bipolar scales are not commonly used for quality assessment. Reference standards
for profiling of wine have been developed (Noble et al., 1987) and these assist in quantitative
tests in terms of defining differences and anchoring unstructured line scales. They are also used
extensively for training of panelists.
Simple descriptive tests are used to quantify a simple well-defined characteristic.
Quantitative descriptive analysis (QDA) is more complex than this, as several different attributes
are profiled at the same time (Lawless & Heymann, 1998; Kilcast, 2003). This method has been
used for the profiling of Brett wines (Wirz et al. 2004), wines at different temperatures spiked
24
with 4-ethylphenol (Cliff & King, 2009), as well as for other taints (Pickering et al., 2008).
Modifications of this method have been used for profiling Brett character in several other
instances (Etiévant et al., 1988; Licker et al, 1999; Ugarte et al., 2005).
Etiévant et al. (1989) performed descriptive analysis on a range of wines, with specific
focus on how 4-ethylphenol and 4-ethylguaiacol affect the sensory profile of the wine. The data
were, however, converted to a ranking and analysed nonparametrically using the Friedman test.
This same method has been employed by other authors (Ugarte et al., 2005). Although this
method compensates for variation between judges, a fair amount of sensory information can be
lost.
In the study by Licker et al. (1999) mentioned in Section X5.3 X, three Cabernet Sauvignon
wines that were identified by their winemaker as having a “no Brett”, “medium Brett” and “high
Brett” character were obtained from the same cellar. QDA was performed on these three wines
in combination with Gas Chromatography-Olfactometry (GC-O) and Charm analysis. Charm
analysis is a technique which attempts to quantify odour intensities in GC-O by means of
repeating the GC-O analysis of a sample at successive dilutions. Judges are expected to
respond when an odour appears, and when an odour stops appearing. A Charm response
chromatogram is produced by plotting the retention index against c, a value related to the
dilution factor and the number of judges that responded to the presence of an odour (c = dn-1,
where d is the dilution factor and n the number of coincident responses). An algorithm then
provides a measure of sensory intensity (Acree et al., 1984).
In the latter study the QDA allowed the researchers to obtain a detailed profile of the
different wines, allowing them to show that there is a distinct increase in plastic, horse sweat
and band-aid™ aromas in the Brett wines, whilst the “no Brett” wine was predominated by fruity,
floral, spicy, earthy and woody aromas. From these data the conclusion was drawn that the
latter aromas were suppressed by Brett character. The combination of this data with the data
obtained from GC-O allowed the researchers to identify a number of different compounds that
contribute to Brett character that had previously not been considered.
7.1.2 45BHedonic tests
The traditional method for the determination of the degree of liking of a wine is the nine-point
hedonic scale. When this method is used, the consumer is asked to indicate which term best
describes his/her attitude towards the products being tasted using the scale with the following
nine categories (Lawless & Heymann, 1998): 9 = Like extremely; 8 = Like very much; 7 = Like
moderately; 6 = Like slightly; 5 = Neither like nor dislike; 4 = Dislike slightly; 3 = Dislike
moderately; 2 = Dislike very much and 1 = Dislike extremely. However, the performance of
preference testing for tainted wines usually takes the form of consumer rejection thresholds, in
which consumers have to identify which of a pair of samples (one containing the taint and one
25
not containing the taint) they prefer more. This method is generally recommended for use with
tainted products (Kilcast, 2003) and has been employed for TCA (Prescott et al., 2005; Teixeira
et al., 2006) and 1,8-cineole (eucalyptol) (Saliba et al., 2009). However, this approach tests
preference between, and not rejection of, a tainted sample, and does therefore not necessarily
provide relevant information. Such tests also ignore the inherent differences that exist between
individual consumers and groups of consumers.
Two non-standard methodologies have been used for hedonic testing, although both
attempt to set up consumer rejection thresholds. A test was done by Eteiévant et al. (1989) to
determine the preferred level of 4-ethylphenol in wine. This was done by adjusting the 4-
ethylphenol content of an existing wine to 8 concentrations ranging from 860 to 3880 µg/L , and
supplying all judges with a reference sample containing an 12% ethanolic solution of 4-
ethylphenol. A 100 mm line scale was used, with “No reference odour” indicated at the left of
the line, and “Intensity of reference odour just right” indicated at 50 mm. Assessors were
instructed to taste the wine and indicate their opinion about the reference flavour level by
marking a point on the scale that best describes their opinion about the wine. In other words, if
an assessor could not detect the reference odour, a mark would be made at 0 mm; if an
assessor could detect the odour but felt that it was present at a too low concentration, a mark
would be made between 0 mm and 50 mm. If the odour was at exactly the concentration that
the assessor preferred, a mark would be made at 50 mm, and if this odour was found to be
excessive, a mark would be made between 50 and 100 mm. A “mean ideal concentration” of 4-
ethylphenol was determined for each assessor, and the overall mean was found to be 1800
µg/L, with the 95% confidence levels indicated to be between 1200 and 2400 µg/L. This test
mixes the determination of detection thresholds with a preference test using a method most
often used for descriptive analysis, which can make it unreliable.
A so-called “limit of preference threshold” was determined by Chatonnet et al. (1992) by
asking 20 trained tasters which of two samples in a subset they preferred; one being spiked with
the reference compound and one being “clean”. The concentrations of the compounds that was
rejected by 50% of the tasters was set as the limit of preference threshold. This threshold was
determined as 620 µg/l for 4-ethylphenol and 140 µg/l for 4-ethylguaiacol. The limit of
preference threshold for 4-ethylphenol and 4-ethylguaiacol combined in a ratio of 10:1 was
found to be 426 µg/L. This value is widely used as a “rejection threshold” for wines containing
volatile phenols. Although this test utilises an accepted methodology, it uses trained tasters,
which are not recommended for preference test (Lawless & Heymann, 1998). The sample size
of 20 judges is also too small to determine preference of a population (Kilcast, 2003). For this
reason, the validity of the value and the common use of this value as rejection criterion should
be questioned.
Although the determination of consumer rejection thresholds is the accepted and
referred method of hedonic testing of tainted wines (Kilcast, 2003; Prescott et al., 2005; Teixeira
26
et al., 2006; Saliba et al., 2009), it may not be the most appropriate method for hedonic testing
of wines tainted with Brett character. This is because Brett character is seen by some people to
add complexity to wines, and therefore some “tainted” wines may be more acceptable to
consumers than non-tainted wines.
7.1.3 46BTesting for association using statistical analyses
Quantitative sensory data can be combined with chemical data by means of multivariate
statistics, which allows for more direct and accurate interpretation of the relationship between
chemical and sensory data (Marias et al., 1979; Noble & Ebeler, 2002; Chung et al., 2003;
Boselli et al., 2004; Genovese et al., 2007). The simplest multivariate method is Principal
Component Analysis (PCA), which reduces multidimensional datasets to orthogonal “principal
components”. These components are modelled in such a way that the first component is in the
direction describing most of the variation in the dataset, and each successive orthogonal
component explains successively less of the variation. An overview of the most important
aspects of complex datasets can therefore be given by plotting principal components against
each other. On a PCA plot, samples are plotted as “scores”, and variables are plotted as
“loadings”. On such a plot, loadings that associate positively are in the same direction, and
loadings that associate negatively are in opposite directions. Loadings that are perpendicular
have no association with one another. Similarly, scores that are similar associate and scores
that are dissimilar associate negatively (Esbensen, 2002). The underlying principles of PCA are
common to many multivariate methods.
When it comes to the combination of sensory and chemical data, two general sets of
multivariate methods are used; namely symmetric and asymmetric methods. Symmetric
methods include General Proscrutes Analysis (GPA) and Canonical Correlation Analysis (CCA)
(Dijksterhuis, 1995). These methods treat both datasets the same, and are more interested in
finding relationships between the different datasets than predicting the scores and loadings of
one dataset from the scores and loading of the other dataset, as is the case with assymetric
methods. For example in GPA, the “one (PCA) space is rotated, reflected, and stretched or
shrunk (scaled) to optimally match the second space” (Noble & Ebeler, 2002). GPA is
considered a suitable method for understanding the relationships between chemical
composition and sensory attributes (Chung et al., 2003).
Asymmetric methods attempt to predict the scores and loading of one dataset from the
scores and loadings of another dataset. Examples of commonly used asymmetric methods are
partial least squares regression methods (PLS, PLS2) and principal component regression
(PCR) (Dijksterhuis, 1995). PLS is a “soft modelling” technique that extracts “factors” or latent
variables. These factors are linear combinations of one set of variables that predict a large
amount of variation in another set of variables (Noble & Ebeler, 2002). PCR, on the other hand,
27
is a multiple linear regression performed on principal component scores (Esbensen, 2002).
Although a discussion of the suitability of these methods are beyond the scope of this work, it
can be noted that PLS appears to be more commonly used than PCR for the modelling of
sensory data (Noble & Ebeler, 2002; Frøst & Noble, 2002; Chung, 2003).
Asymmetric multivariate methods form the basis of electronic nose and electronic tongue
technologies (Buratti et al., 2007), which have also been investigated for identification of Brett
character in wine (Cyncar et al., 2007). This method is discussed in Section X7.2 X.
Preference mapping is a method that can be used to correlate the preference of different
groups of consumers to the sensory and/or chemical qualities of food products and beverages
(Tuorila & Monteleone, 2009). Two basic types of preference mapping exist, namely internal
preference mapping and external preference mapping (Anon, 2009). In their review on
preference mapping techniques, Meilgaard et al. (2007), however, designates PLS mapping to
be a third type of preference map. Internal preference mapping (the MDPREF procedure) is
used to summarise the liking of a large group of consumers to various products, without taking
any of the intrinsic properties of the products into account. External preference mapping (the
PREFMAP procedure) involves relating the preference of consumers to specific characteristics
of products like sensory properties, physio-chemical properties or economic attributes (Anon,
2009). PCA is the underlying multivariate technique for both these methods. PLS mapping is a
direct application of partial least squares regression (PLS-2), with the sensory or chemical data
in the X space, and the consumer data in the Y (predictive) space (Meilgaard et al., 2007).
Preference mapping has been used for wine (Frøst & Noble, 2002), several food
products (apples (Thybo et al., 2003), yogurt, pudding (Elmore et al., 1999) and beverages (tea
(Cho et al., 2005) coffee (Geel et al., 2005), and rice wine (Lee & Lee, 2008)). This method has
not yet been used for the analysis of wine taints, but has immense potential for application to a
problem such as Brett character.
7.2 17BChemical methodologies for analysing compounds associated with Brett
character
It is essential to accurately determine the concentrations of the relevant compounds associated
with Brettanomyces in any study attempting to study this phenomenon in detail.
The most common method for the determination of the ethylphenols is gas
chromatography, which can be coupled to either a flame ionization detector (FID) (Martorell et
al., 2002; Monje et al., 2003), mass spectrometry (MS) (Etiévant, 1981) or tandem mass
spectrometry (MS-MS) (Pizarro et al., 2007). Mass spectrometry appears to be more popular
than FID. Tandem mass spectroscopy provides even better sensitivity, but the instrumentation
is more expensive and not as commonly available.
28
In terms of sample preparation, liquid-liquid extraction (LLE) is frequently used. Solvents
include dichloromethane (Chatonnet et al., 1992; 1993), diethyl ether-pentane (Pollnitz et al.,
2000) or ether-hexane (Rodriguez et al. 2001; Dias et al. 2003; Martorell et al., 2006).
Dispersive liquid-liquid microextraction (DLLME) has also been investigated as an alternative
sample preparation method (Fariña et al, 2007). This method uses two solvents – an extractor
solvent which has a higher density than the sample and has an affinity for the compounds in
question, and a disperser solvent which is miscible with both the sample and the extractor
solvent. The disperser solvent acts as a “carrier” between the sample and the extractor solvent.
An example of such a solvent system is carbon tetrachloride (extractor solvent) and acetone
(disperser solvent). Although this method produced relatively high limits of quantification and
detection (see XTable 2.3 X), its small sample volume, short extraction time (6 minutes) and low
solvent usage, all which make it less expensive than LLE, are major advantages.
Stir bar sorptive extraction has also been investigated (Díez et al., 2004), but the limits of
detection and quantification reported were significantly higher than many other methods, and, in
the case of 4-ethylguaiacol (159 µg/L) much higher than the sensory detection threshold (33
µg/L). This method also has a rather extensive extraction time (60 minutes).
Significant research has been done into the use of sorptive extraction methods for the
analysis of ethylphenols. These include headspace solid-phase microextraction (HS-SPME)
(Monje et al., 2002; Martorell et al., 2002; Carrilo et al., 2006; Botou & Chatonnet, 2007) and
multiple headspace microextraction (Pizarro et al., 2007). A general advantage of these
methods is that they require little sample preparation and do not require a solvent. A common
choice of fibre is divinylbenzene-caboxen-poly(dimethylsiloxane) (DVD/CAR/PDMS) (Carillo et
al., 2006; Carillo & Tena, 2007; Botou & Chatonnet, 2007), although polydimethylsiloxane
(PDMS) (Martorell et al., 2002) and polyacrylate (Monje et al., 2002) have also been used with
success. Pizarro et al. (2007) reported severe matrix effects for MS-SPME, which may have
been due to the use of a carbowax/divinylbenzene fibre. HS-SPME coupled to GC-MS also
generally gives low LOD’s and LOQ’s (see XTable 2.3 X). In most cases, however, matrix effects
are significant, which means that standard addition must be used for quantification, which is
extremely time-consuming. (MHS-SPME does not have this disadvantage (Pizarro et al., 2007),
but requires several extraction steps, which is also time-consuming.)
Derivitisation prior to extraction has also been applied. Such a derivitisation step is
necessary for the determination of 4-ethylcatechol, as it highly polar and non-volatile, and
causes peak tailing even when a polar GC column is used. This is overcome by for example
acetylation of the compound through the addition of acetic anhydride (Carillo & Tena, 2007).
Although this method is successful for the detection of 4-ethylcatechol, its limits of quantification
an detection are approximately double those found using similar methods (Carillo et al., 2006).
29
Table 2.3. Recent methods used for determination of ethylphenols in wines showing limits of
detection (LOD) and limits of quantification (LOQ). All values are in µg/L.
Method LOD
4-EP
LOD
4-EG
LOD
4-EC
LOQ
4-EP
LOQ
4-EG
LOQ
4-EC Authors
DLLMEa GC-MS 44 28 - 147 95 - Fariña et al., 2007
SBSEb GC-MS 6 159 - 21 529 - Díez et al., 2004
HS-SPMEc GC-FID 2 1 - 5 5 - Martorell et al., 2002
HS-SPME GC-MS 7 1 - 15 2 - Carillo et al., 2006
HS-SPME GC-MS 11.5 3.8 - 25.1 9.1 - Botou & Chatonnet, 2007
HS-SPME GC-MS 17 2 4 30 3 6 Carillo & Tena, 2007
MHS-SPMEd GC-MS/MS 0.06 0.06 - 0.20 0.18 - Pizarro et al., 2007
LC-MS/MS 10 10 - 50 50 - Caboni et al., 2007
HPLC-DAD 10 10 - 50 50 - Caboni et al., 2007
HPLC- fluorescence 1 10 - 5 50 - Caboni et al., 2007
HPLC-CEADe 1.30 1.57 - 2.59 3.13 - Larcher et al., 2007
HPLC-CEAD 1.30 1.57 0.33 2.59 3.13 1.1 Larcher et al., 2008 a Dispersive liquid-liquid microextraction b Stir bar sorptive extraction c Head-space solid-phase microextraction d Multiple head-space solid-phase microextraction e Coulometric array detector
There has also been recent interest in liquid chromatographic methods for the
determination of the ethylphenols. These methods require no sample preparation and have the
significant advantage that they are more suitable for the determination of 4-ethylcatechol than
GC-MS. A HPLC-Coulometric method for the determination of 4-ethylphenol and 4-
ethylguaiacol was developed by Larcher et al. (2007), which was subsequently adapted for the
analysis of 4-ethylcatechol (Larcher et al., 2008). As can be seen in XTable 2.3 X, these methods
have relatively low limits of detection and quantification, especially when compared to GC-MS
preceded with derivitisation (Carillo & Tena, 2007).
Caboni et al. (2007) developed a LC-MS/MS method as well as a HPLC-diode array
method and a HPLC-fluorescence method. However, the limits of quantification for these
methods are high when compared to GC-MS, and especially high when compared to GC-MS-
MS (Pizarro et al., 2007). The main advantage of this method is therefore that no sample
preparation is required. An HPLC-DAD method has also recently been developed for the
analysis of 4-vinylcatechol, the precursor of 4-ethylcatechol (Hisomoto et al., 2009). The other
ethylphenols could also be detected with this method.
Cyncar et al. (2007) tested the feasibility of using an MS electronic nose method to
discriminate between commercial wines according to their 4-ethylphenol levels. The categories
of high (higher than 500 μg/L), medium (between 500 and 200 μg/L), and low (lower than 100
μg/L) were tested and the method was found to have a classification rate of 67%. This method
might be able to produce at least a 40% reduction in cost as compared to headspace GC-MS or
sensory analysis methods. The method also has potential as a routine method for wine quality
monitoring. Further studies on this instrument (Berna et al., 2008) found that concentrations of
30
4-ethylphenol of higher than 20 µg/L could be reliably estimated. In spite of this improvement,
this technique is not developed to a level where it is applicable for routine wine monitoring.
8 SUMMARY
Brettanomyces forms part of the incredibly complex microbiology of wine. Although this micro-
organism has been known for several decades, there are still several unanswered questions
regarding its sensory effects in red wine, particularly South African red wine. The first of these is
what the potential effect of elevated levels of 4-ethylcatechol has in South African wines, and
what the effects of other metabolites of Brettanomyces are on the sensory character of wine.
Although attempts have been made to study these effects, more accurate answers could be
obtained by using appropriate sensory research methodology, as well as combining sensory
data with chemical data through multivariate statistics. A final area of question is how the South
African wine consumer responds to wines spoiled by Brettanomyces, and whether or not these
consumers find wine spoiled by this micro-organism objectionable.
The answers to these questions all relate to the chemical diagnosis of the sensory effect
which is Brett character. The question remains at which levels the Brett-related spoilage
compounds can be detected in wine, as well as at which levels these compounds become
objectionable. Furthermore, the sensory effect of 4-ethylcatechol and its sensory interaction with
the other Brett-related spoilage compounds has not yet been investigated. These two aspects
are of utmost importance for the South African wine industry for several reasons. 4-
ethylcatechol is not currently considered as an important Brett compound, and diagnostic
analyses do not include this compound. However, Pinotage, a uniquely South African cultivar,
contains exceedingly high quantities of caftaric acid and caffeic acid, making this cultivar more
susceptible to high levels of this compound.
9 REFERENCES
Acree, T. E., Barnard, J. & Cunningham, D. G. (1984) A procedure for the sensory analysis of
gas chromatographic effluents. Food Chemistry. 14, 273 - 286.
An, S.R., Shim, S.D., Lee, J.Y., Hong, K. W & Lee, S. J. (2009) Investigation on beef quality
Indicator of off-flavor development during storage (abstract only). Korean Journal for
Food Science of Animal Resources. 29, 325 - 333.
Anomynous (2009). What is Preference Mapping? [WWW document].
HUhttp://www.xlstat.com/en/support/tutorials/prefmap.htm. 29 October 2009UH.
31
Arriagada-Carrazana, J. P., Sáez-Navarette, C. & Bordeu, E. (2005) Membrane filtration effects
on aromatic and phenolic quality of Cabernet Sauvignon wines. Journal of Food
Engineering. 68, 363 - 368.
ASTM (1997). Standard practice E 679-91. Determination of odor and taste thresholds by an
ascending concentration series method of limits. Philadelphia, USA: American Society
for Testing and Materials
Atanasova, B., Thomas-Danguim, T., Langious, D., Nickluas, S., Chabanet, C., & Etiévant, P.
(2005) Perception of wine fruity and woody notes: influence of peri-threshold odorants.
Food Quality and Preference. 16, 504 - 510.
Barata, A., Caldiera, R., Botelheiro, R., Pagliara, D., Malfeito-Ferreira, M. & Loureiro, V. (2008)
Survival patterns of Dekkera bruxellensis in wines and inhibitory effects of sulphur
dioxide. International Journal of Food Microbiology. 121, 201 - 207.
Barrera-García, V. D., Gougeon, R. D., Voilley, A. & Chassagne, D. (2006) Sorption behavior of
volatile phenols at the oak wood/wine interface in a model system. Journal of Agricultural
and Food Chemistry. 54, 3982 - 3989.
Berna, A. Z., Trowell, S., Cynkar, W. & Cozzolino, D. (2008) Comparison of metal oxide-based
electronic nose and mass spectrometry-based electronic nose for the prediction of red
wine spoilage. Journal of Agricultural and Food Chemistry. 56, 3238 - 3244.
Boselli, E., Boulton, R. B., Thorngate, J. H. & Frega, N. G. (2004) Chemical and sensory
characterisation of DOC red wines from Marche (Italy) related to vintage and grape
cultivars. Journal of Agricultural and Food Chemistry. 52, 3843 - 3854.
Boutou, S. & Chatonnet, P. (2007) Rapid headspace solid-phase microextraction/gas
chromatographic/mass spectrometric assay for the quantitative determination of some of
the main odorants causing off-flavours in wine. Journal of Chromatography A. 1141, 1 -
9 .
Brown, R. C., Sefton, M. A., Taylor, M. A., Taylor, D. K. & Elsey, G. M. (2006) An odour
detection threshold determination of all four possible stereoisomers of oak lactone in a
white and red wine. Australian Journal of Grape and Wine Research. 12, 115 - 118.
Buratti, S., Ballabio, D., Benedetti, S. & Cosia, M. S. (2007) Prediction of Italian red wine
sensorial descriptors from electronic nose, electronic tongue and spectrophotometric
measurements by means of Genetic Algorithm regression models. Food Chemistry. 100,
211 - 218.
Caboni, P., Sarais, G., Cabras, M. & Agoini, A. (2007) Determination of 4-ethylphenol and 4-
ethylguiacol in wine by LC-MS-MS and HPLC-DAD-fluorescence. Journal of Agricultural
and Food Chemistry. 55, 7288 - 7293.
Campolongo, S., Rantsiou, K., Giordano, M., Gerbi, V. & Cocolin, L. (2009) Brettanomyces
bruxellensis: incidence and diversity in Italian Wines as determined by molecular
Methods. American Journal of Enology and Viticulture. 60, 398.
32
Carillo, J. D., Garrido-Lόpez, A. & Tena, M. T (2006) Determination of volatile oak compounds in
wine by headspace solid-phase microextraction and gas chromatography-mass
spectrometry. Journal of Chromatography A. 1102, 25 - 36.
Carrillo, J. D. & Tena, M. T. (2007) Determination of ethylphenols in wine by in situ derivitisation
and headspace solid-phase microextraction-gas chromatography-mass spectrometry.
Analytical and Bioanalytical Chemistry. 387, 2547 - 2558.
Chassagne, D.M., Cuilloux-Benatier, M., Alexandre, H. & Voilley, A. (2005) Sorption of wine
volatile phenols by yeast lees. Food Chemistry. 91, 39 - 44.
Chatonnet, P., Dubourdieu, D. & Boidron, J. (1995) The influence of Brettanomyces/Dekkera sp.
yeasts and lactic acid bacteria on the ethylphenol content of red wines. American
Journal of Enology and Viticulture. 46, 463 - 468.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Lavigne, V. (1993) Synthesis of volatile phenols by
Saccharomyces cerivisae in Wines. Journal of the Science of Food and Agriculture. 62,
191 - 202.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
Cho, H., Chung, S., Kim, H. & Kim, K. (2005) Effect of sensory characteristics and non-sensory
factors on consumer liking of various canned tea products. Journal of Food Science. 70,
532 - 538.
Chung, S., Heymann, H. & Grün, I. U. (2003) Application of GPA and PLSR in correlating
sensory and chemical data sets. Food Quality and Preference. 14, 485 - 495.
Ciani, M. & Ferraro, L. (1997) Role of oxygen on acetic acid producing by
Brettanomyces/Dekkera in winemaking. Journal of the Science of Food and Agriculture.
75, 489 - 495.
Ciani, M., Maccarelli, F. & Fatichenti, F. (2003) Growth and fermentation behavior of
Brettanomyces/Dekkera yeast under different conditions of aerobiosis. World Journal of
Microbiology and Biotechnology. 19, 419 - 422.
Cliff, M. A. & King, M. C. (2009) Influence of serving temperature and wine type on perception of
ethyl acetate and 4-ethyl phenol in wine. Journal of Wine Research. 20, 45 - 52.
Conterno, L., Hesford, F. & Henick-Kling, T. (2005) Volatile phenols off-odor production by
different strains of Brettanomyces bruxellensis. American Journal of Enology and
Viticulture. 56, 292A.
Cyncar, W., Cozzolino, D., Dambergs, B., Janik, L. & Gishen, M. (2007) Feasibilty study on the
use of a head space mass spectrometry electronic nose (MS e_nose) to monitor red
wine spoilage induced by Brettanomyces yeast. Sensors and Actuators. 124, 167 - 171.
De Villiers, A., Majek, P., Lynen, F., Crouch, A., Lauer, H. & Sandra, P. (2005) Classification of
South African red and white wines according to grape variety based on the non-coloured
phenolic content. European Food Research and Technology. 221, 520 - 528.
33
Dias, L., Pereira-da-Silva, S., Tavares, M., Malfeito-Ferreira, M. & Loureiro, V. (2003) Factors
affecting the production of 4-ethylphenol by the yeast Dekkera bruxellensis in enological
conditions. Food Microbiology. 20, 377 - 384.
Díez, J., Domínguez, C., Guillén, D. A., Veas, R. & Barosso, C. G. (2004) Optimisation of stir
bar sorptive extraction for the analysis of volatile phenols in wines. Journal of
Chromatography A. 1025, 263 - 267.
Dijksterhuis, G. (1995) Multivariate data analysis in sensory and consumer science: An
overview of developments. Trends in Food Science & Technology. 6, 206 - 211.
du Toit, W. J., Pretorius, I. S. & Lonvaud-Funel, A. (2005) The effect of sulphur dioxide and
oxygen on the viability and culturabilty of a strain of Acetobacter pasteurianus and a
strain of Brettanomyces bruxellensis isolated from wine. Journal of Applied Microbiology.
98, 862 - 871.
Elmore, J. R., Heymann, H., Johnson, J. & Hewett, J. E. (1999) Preference mapping: relating
acceptance of "creaminess" to a descriptive sensory map of a semi-solid. Food Quality
and Preference. 10, 465 - 475.
Esbensen, K. H. (2002) Multivariate Data Analysis, 5th ed. Osla, Norway: CAMO Process AS
Escudero, A., Campo, E., Fariña, L., Cacho, J. & Ferreira, V. (2007) Analytical characterization
of the aroma of five premium red wines. Insights into the role of odor families as the
concept of fruitiness of wines. Journal of Agricultural and Food Chemistry. 55, 4501 -
4510.
Eteiévant, P. X. (1981) Volatile phenol determination in wine. Journal of Agricultural and Food
Chemistry. 29, 65 - 67.
Eteiévant, P. X., Issanchou, S. N., Marie, S., Ducruet, V. & Flanzy, C. (1989) Sensory impact of
volatile phenols on red wine aroma: influence of carbonic maceration time and storage.
Sciences des aliments. 9, 19 - 31.
Farina, L., Boido, E., Carrau, F. & Dellacassa, E. (2007) Determination of volatile phenols in red
wines by dispersive liquid-liquid microextraction and gas chromatography-mass
spectrometry detection. Journal of Chromatography A, 1157. 1157, 46 - 50.
Ferreira, V., Jaruta, I. & Cacho, J. (2006) Physiochemical model to interpret the kinetics of
aroma extraction during wine aging in wood. Model limitations suggest the necessary
existance of biochemical processes. Journal of Agricultural and Food Chemistry. 54,
3047 - 3054.
Ferreira, V., Lόpez, R. & Cacho, J. F. (2000) Quantitative determination of the odorants of
young red wines from different grape varieties. Journal of the Science of Food and
Agriculture. 80, 1659 - 1667.
Frøst, M. B. & Noble, A. C. (2002) Preliminary study on the effect of knowledge and sensory
expertise on liking for red wines. American Journal of Enology and Viticulture. 53, 275 -
284.
34
Fugelsang, K. C. & Zoeklein, B. W. (2003) Population dynamics and effects of Brettanomyces
bruxellensis strains on Pinot noir (Vitis vinifera L.) wines. American Journal of Enology
and Viticulture. 54, 294 - 300.
Geel., L., Kinnear, M. & de Kock, H.L. (2005) Relating consumer preferences to sensory
attributes of instant coffee. Food Quality and Preference. 16, 237 - 244.
Gerbaux, V., Vincent, B. & Bertrand, A. (2002) Influence of maceration temperature and
enzymes on the content of volatile phenols in Pinot noir wines. American Journal of
Enology and Viticulture. 53, 131.
Harwood C. S. & Canale-Parola, E. (1981) Adenosine-5'-triphosphate-yielding pathways of
branched-chain amino acid fermentation by a marine spirochete. Journal of Bacteriology.
148, 117 - 123.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11-13.
Hesford, F., Schneider, K., Porret, N. & Gafner, J. (2004) Identification and analysis of 4-ethyl
catechol in wines tainted by Brettanomyces off-flavor. American Journal of Enology and
Viticulture. 55, 304A.
Hisomoto, M., Okuda, T., Nishimoto, S., Tani, K., Tachibana, M., Koizumi, H., Kiba, N. &
Yokotsuka, K. (2009) Determination of 4-vinylcatechol in wine by HPLC-DAD coupled
with flourescence detection. American Journal of Enology and Viticulture. 60, 402.
Jiménez-Moreno, N. & Ancín-Azpilicueta, C (2009) Sorption of volatile phenols by yeast cell
walls. Internationl Journal of Wine Research. 1, 11-18.
Kelly, J. (2003) What is Brettanomyces and why are so many winemakers concerned about it?
The Australian & New Zealand Grapegrower & Winemaker. 474, 54 - 57.
Kilcast, D. (2003) Sensory analytical methods in detecting taints and off-flavours in food. IN:
Taints and off-flavours in foods. (edited by B. Baigrie). Pp 5 - 30. Cambridge, England:
Woodhead Publishing Ltd.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Larcher, R., Nicolini, G., Puecher, C., Bertoldi, D., Moser, S. & Favaro, G. (2007) Determination
of volatile phenols in wine using high-performace liquid chromatography with a
coulometric array detector. Analytica Chimica Acta. 582, 55 -60.
Lawless, H. T. & Heymann, H. (1998). Sensory evaluation of food, principles and practice.
New York, USA: Chapman and Hall.
Le Berre, E., Atanasova, B., Langloi, Etiéviant, P. & Thamas-Danguin, T. (2007) Impact of
ethanol on the perception of wine odorant mixtures. Food Quality and Preference. 18,
901 - 908.
35
Lee, S. J. & Lee, K. G. (2008) Understanding consumer preferences for rice wines using
sensory data. Journal of the Science of Food and Agriculture. 88, 690 - 698.
Licker, J. L, Acree, T. E. & Henick-Kling, T. (1999) What is "Brett" (Brettanomyces) flavour? A
preliminary investigation. In: Chemistry of Wine Flavour. ACS Symposium Series (edited
by A. L. Waterhouse & S. E. Ebeler), Pp 96 - 115. Washington DC: American Chemical
Society
Loureiro, V. & Malfeito-Ferreira, M. (2003) Spoilage yeasts in the wine industry. International
Journal of Food Microbiology. 86, 25 - 20.
Mansfield, A. K., Zoecklein, B. W. & Whinton, R. S. (2002) Quantification of glycosidase activity
in selected strains of Brettanomyces bruxellensis and Oenococcus oeni. American
Journal of Enology and Viticulture. 53, 303 - 307.
Marais, J., Van Rooyen, P. C. & Du Plessis, C. S. (1979) Objective quality rating of Pinotage
wine. Vitis. 18, 31 - 39.
Martineau, B., Acree, T. E. & Henick-Kling, T. (1995) Effect of wine type on detection threshold
for diacetyl. Food Research International. 28, 139 - 143.
Martorell, N., Martí, M. P., Mestres, M., Busto, O. & Guasch, J. (2002) Determination of 4-
ethylguaiacol and 4-ethylphenol in red wine using headspace-solid-phase
microextraction-gas chromatograpgy. Journal of Chromatography A. 978, 349 - 354.
Martorell, P., Barata, A., Malfeito-Ferreira, M., Fernández-Espinar, M. T., Loureiro, V. & Querol,
A. (2006) Molecular typing of the yeast species Dekkera bruxellensis and Pichia
guilliermondii recovered from wine related sources. International Journal of Food
Microbiology. 106, 79 - 84.
Mazzoleni, V. & Maggi, L. (2007) Effect of wine style on the perception of 2,4,6-trichloroanisole,
a compound related to cork taint in wine. Food Research International. 40, 649 - 699.
Millet, V. & Lonvaud-Funel, A. (2000) The viable but non-culturable state of wine micro-
organisms during storage. Letters in Applied Microbiology. 30, 136 - 141.
Monje, M., Privat, C., Gastine, V & Nepveu, F. (2002) Determination of ethylphenol compounds
by headspace solid-phase microextraction in conjunction with gas chromatography and
flame ionization detection. Analytica Chimica Acta. 458, 111 -117.
Noble, A. C. & Ebeler, S. E. (2002) Use of Multivariate statistics in understanding wine flavour.
Food Reviews International. 18, 1 - 21.
Noble, A.C., Arnold, R.A., Buechsenstein, J., Leach, E. J., Schmidt, J. O. & Stern, P. M. (1987)
Modification of a standardized system of wine aroma terminology. American Journal of
Enology and Viticulture. 38, 143 - 146.
Norris, L. (2004) Unraveling the mystery of Brettanomyces flavor. American Journal of Enology
and Viticulture. 55, 304A.
36
Oelofse, A., Pretorius, I. S., & du Toit, M. (2008) Significance of Brettanomyces and Dekkera
during winemaking: A synoptic review. South African Journal of Enology and Viticulture.
29, 128 - 144.
O'Mahony, M. (1995) Who told you the triangle test was simple?. Food Quality and Preference.
6, 227 - 738.
Parish, M., Kelly, M. & Baldwin, G. (2003) Managing Brett. Australian and new Zealand Wine
industry Journal. 18, 12 - 15.
Pickering, G. J., Karthik, A., Inglis, D., Sears, M. & Ker, K. (2007) Determination of ortho- and
retronasal detection thresholds for 2-isopropyl-3methoxypyrazine in wine. Journal of
Food Science. 27, 468 - 472.
Pickering, G. J., Spink, M., Kotserdis, Y., Inglis, D., Brindle, I. D., Sears, M. & Beh, A. L (2008)
Yeast strain affects 3-isopropyl-2-methoxypyrazine concentration and sensory profile in
Cabernet Sauvignon wine. Australian Journal of Grape and Wine Research. 14, 230 -
237.
Pizarro, C., Pérez-del-Notario, N. & González-Sáiz, J. M. (2007) Determination of Brett
character responsible compounds in wines by using multiple headspace solid-phase
microextraction. Journal of Chromatography A. 1143, 176 - 181.
Pollnitz, A. P., Pardon, K. H. & Sefton, M. A. (2000) Quantitative analysis of 4-ethylphenol and
4-ethylguaiacol in red wine. Journal of Chromatography A. 874, 101 - 109.
Prescott, J., Norris, L., Kunst, M. & Kim, S. (2005) Estimating a "consumer rejection threshold"
for cork taint in white wine. Food Quality and Preference. 16, 345 - 349.
Rayne, S., & Eggers, N. (2007) Volatile phenols from Brettanomyces in barreled red wines: are
they stable? And where do they reside? The Australian & New Zealand Grapegrower &
Winemaker. 522, 64 - 69.
Renouf, V., Louvaud-Funel, A. & Coulon, J. (2007) The origin of Brettanomyces bruxellensis in
wines: A review. Journal Internationale de Science de Vigne et Vin. 41, 161 - 173.
Renouf, V., Perello, M., de Revel, G. & Lonvaud-Funel, A. (2007) Survival of wine
microorganisms in the bottle during storage. American Journal of Enology and
Viticulture. 58, 379 - 386.
Robinson, K. M., Klien, B. P. & Lee, S. -Y. (2005) Utilizing the R-index measure for threshold
testing in model caffeine solutions. Food Quality and Preference. 16, 283 - 289.
Rodrigues, N., Gonçalves, G., Pereira-da-Silva, S., Malfeito-Ferreira, M. & Loureiro, V. (2001)
Development and use of a new medium to detect yeasts of the genera
Dekkera/Brettanomyces. Journal of Applied Microbiology. 90, 588 - 599.
Romano, A., Perello, M. C., de Revel, G. & Lonvaud-Funel, A. (2008) Growth and volatlie
compound production by Brettanomyces/Dekkera bruxellensis in red wine. Journal of
Applied Microbiology. 104, 1577 - 1585.
37
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of “Brett character”. Food Chemistry. 114, 15 - 19.
Rossouw, M. & Marias, J. (2004) The phenolic composition of South African Pinotage, Shiraz
and Cabernet Sauvignon wines. South African Journal of Enology and Viticulture. 25, 94
- 104.
Saliba, A. J., Bullock, J. & Hardie, W. J. (2009) Consumer rejection threshold of 1,8-cineole
(eucalyptol) in Australian red wine. Food Quality and Preference. 20, 500 - 504.
Saurez, R., Saurez-Lepe, J. A., Morata, A. & Calderon, F. (2007) The production of
ethylphenols in wine by yeasts of the genera Brettanomyces and Dekkera: A review.
Food Chemistry. 102, 10 - 21
Shinohara, T., Kubodera, S. & Yanagida, F. (2000) Distrubution of phenolic yeasts and
production of phenolic off-flavours in wine fermentation. Journal of Bioscience and
Bioengineering. 90, 90 - 97.
Silva, P., Cardoso, H. & Gerós, H. (2004) Studies on the wine spoilage capacity of
Brettanomyces/Dekkera spp. American Journal of Enology and Viticulture. 55, 65 - 72.
Swiegers, J. H., Bartowsky, R. J., Henschke, P. A. & Pretorius, I. S. (2005) Yeast and bacteria
modulation of wine aroma and flavour. In: AWRI – Advances in wine science (edited by
R. J. Blair, M. E. Francis & I. S. Pretorius A. L.), Pp 159 - 187. Glen Osmond, Australia:
Australian Wine Research Institute
Teixeira, M. I. V., Romão, M. V. S., Broze, M. R. & Boas, L. V. (2006) 2,4,6-Trichloroanisol: A
consumer panel evaluation. Ciência e Técnica Vitivinícola. 2, 53 - 65.
Thybo, A. K., Kühn, B. F. & Martens, H. (2003) Explaining Danish children's preferences for
apples using instrumental, sensory and demographic/behavioural data. Food Quality and
Preference. 15, 53 - 63.
Tuorila, H., Monteleone, E. (2009) Sensory food science in the changing society: Oppurtunities,
needs, and challenges. Trends in Food Science & Technology. 20, 64 - 62.
Ugarte, P., Agosin, E., Bordeu, E. & Villalobos, J. I. (2005) Reduction of 4-ethylphenol and 4-
ethylguaiacol concentration in red wines using reverse osmosis and adsorption.
American Journal of Enology and Viticulture. 56, 30 - 36.
Uscanga, M., Delia, M. & Strehaiano, P. (2000) Nutritional requirements of Brettanomyces
bruxellensis: Growth and physiology in batch and chemostat cultures. Canadian Journal
of Microbiology. 46, 1046 - 1050.
Vichi, S., Romero, A., Tous, J., Tamames, E. L. & Buxaderas, S. (2009) Determination of
volatile phenols in virgin olive oils and their sensory significance. Journal of
Chromatography A. 1211, 1 - 7.
Wirz, D. O., Heymann, H. & Bisson, L. F. (2004) Descriptive analysis of Brettanomyces-infected
Cabernet Sauvignon wines. American Journal of Enology and Viticulture. 55, 303A.
38
Wood, C., Siebert, T. E., Parker, M., Capone, D. L., Elsey, G. M., Pollnitz, A. P., Eggers, M.,
Meier, M., Vössing, T., Widder, S., Krammer, G., Sefton, M. A. & Herderich, M. J. (2008)
From wine to pepper: Rotundone, an obscure sesquiterpene, is a potent spicy aroma
compound. Journal of Agricultural and Food Chemistry. 56, 3738 - 3744.
Wright, J. M. & Parle, J. N. (1974) Brettanomyces in the New Zealand wine industry. New
Zealand Journal of Agricultural Research. 17, 273 - 278.
39
Chapter 3: 4BThe determination of detection thresholds of eight Brettanomyces-related compounds in Pinotage red wine using two
calculation methods
1 INTRODUCTION............................................................................................................. 40
2 MATERIALS AND METHODS........................................................................................ 43
2.1 Samples....................................................................................................................... 43
2.2 Determination of detection threshold levels............................................................ 44
2.2.1 Subjects and training ............................................................................................. 45
2.2.2 Determination of detection thresholds................................................................. 46
2.2.3 Analysis of data ...................................................................................................... 47
3 RESULTS AND DISCUSSION ....................................................................................... 49
3.1 Comparison between different calculation methods .............................................. 49
3.2 Comparison to literature............................................................................................ 55
4 CONCLUSIONS.............................................................................................................. 57
5 REFERENCES................................................................................................................ 58
40
1 INTRODUCTION
Brett character is a wine spoilage defect which is associated with an unpleasant aroma which is
commonly described as “horse-sweat” or “barnyard”. This defect most commonly occurs in red
wine (Du Toit et al., 2005). It is caused by spoilage by the yeast Brettanomyces, and its
sporulating form, Dekkera (Loureiro & Malfeito-Ferreira, 2003). Sensory descriptors for wines
with Brett character include rancid, band-aid™, soy, horsey, leather, tobacco and putrid (Wirz et
al., 2004). Brett character also masks inherent fruitiness in wines, as well as the varietal
character (Licker et al., 1999; Fugelsang & Zoecklein, 2003; Fariña et al., 2007).
The two compounds most commonly associated with Brett character are the
ethylphenols, 4-ethylphenol and 4-ethylguaiacol. These compounds were linked to the genus
Brettanomyces in the early 1990’s (Chatonnet et al., 1992; 1993) and are generally considered
“markers” for Brettanomyces spoilage. For this reason, these compounds are the subject of
most routine tests for this microorganism in wine. A third ethylphenol, 4-ethylcatechol, was
recently linked to Brettanomyces (Hesford et al., 2004, Hesford & Schneider, 2004, Larcher et
al., 2008).
Isovaleric acid, a short-chain branched fatty acid, has also been linked to
Brettanomyces, but not without controversy. Some authors have found a strong link between
this compound and Brett character (Licker et al., 1999; Fugelsang & Zoecklein, 2003), whereas
other authors have found poor correlations between elevated levels of isovaleric acid and Brett
character (Henske et al., 2004). A recent study by Romano et al. (2009) found a strong link
between the production of ethylphenols and high levels of both isovaleric acid and isobutyric
acid. This implies the production of these two compounds by Brettanomyces, which may lead to
their involvement in sensory interactions in terms of Brett character.
Two vinylphenols, 4-vinylphenol and 4-vinylguaiacol, are also associated with Brett
character, as these compounds are both the precursors (Chatonnet et al., 1992, 1993) and the
breakdown products (Rayne & Eggers, 2007) of ethylphenols. The conversion of the
vinylphenols to the ethylphenols is facilitated by vinylphenol reductase (Chatonnet et al., 1995),
whereas the breakdown of the ethylphenols to vinylphenols is catalysed by p-cresol
methylhydroxylase (PCMH) (Rayne & Eggers, 2007). However, these compounds are produced
by numerous other wine micro-organisms in addition to Brettanomyces. Brettanomyces also
produces varying amounts of acetic acid, which depends on the availability of oxygen to the
organism (Du Toit et al., 2005). Acetic acid, as in the case of the vinylphenols, can also be
produced by several other wine micro-organisms.
A detection threshold can be defined as the lowest concentration at which a compound
can be detected (but not necessarily recognised) by the senses. Although detection thresholds
are usually either olfactory (detected by the sense of smell) or taste thresholds, the use of the
sense of touch (for example for the determination of the detection threshold of a skin irritant) is
41
not impossible. Similarly, the recognition threshold of a compound is the level at which it can be
recognised by the senses. In other words, the detection threshold is the level at which a
difference can be detected, but the nature of the difference is not clear, whereas the recognition
threshold is the level at which the difference can be recognised. This study focuses on the
former. The detection thresholds for the above-mentioned compounds have been determined,
but not necessarily in wine.
Chatonnet et al. (1992) determined the detection threshold of 4-ethylphenol and 4-
ethylguaiacol in both water and model solution. These thresholds were defined as the minimum
concentration under which 50% of 70 panellists failed to taste the difference from a control. The
same authors also determined a “recovery threshold” in red wine by means of a triangular
directional test, which was also defined as the level at which 50% of 70 panellists could correctly
identify a sample containing the compound. These thresholds were 605 µg/L and 110 µg/L for
4-ethylphenol and 4-ethylguaiacol, respectively. The results found during this research
coincidentally correspond to a significance level of less than 5% when evaluated using the
statistical tables for difference testing (Roessler et al., 1987). However, using a chosen
percentage of total panellists (50% in this case) as a basis instead of the statistical tables is still
considered arbitrary and empirical (Lawless & Heymann, 1998). The methodology used by
Chatonnet et al. (1992) is therefore still considered questionable.
The Australian Wine Research Institute (AWRI) undertook an investigation into the
sensory attributes of Brett character in red wine, which included the determination of detection
thresholds of 4-ethylphenol, 4-ethylguaiacol and 4-ethylcatechol (Curtin et al., 2008). These
values were determined in three wines, namely a “neutral” wine, an “oaky” wine and a “green”
wine. It was found that detection thresholds of these compounds are generally significantly
higher in “oaky” wines than in neutral wines, and slightly higher in “green” wines than neutral
wines. For example, the detection threshold of 4-ethylphenol was 368 µg/L in the neutral wine,
425 µg/L in the “green” wine and 569 µg/L in the “oaky” wine. The detection thresholds in the
neutral wine were 158 and 774 µg/L for 4-ethylguaiacol and 4-ethylcatechol, respectively. The
values for these two compounds in the “green” and “oaky” wines were not given. No details of
the method used for the determination of detection thresholds were however included in the
publication.
Apart from the investigations by the AWRI, the detection threshold of 4-ethylcatechol has
been determined in two other reports. Hesford and Schneider (2004) defined this threshold as
the level at which a difference could be tasted by their panel, and found the value to be 60 µg/L.
However, no detail is given about the method used or the statistical basis for the test, and the
validity of this value is therefore questionable. More recently, Larcher et al. (2008) used a duo-
trio test and 5% significance levels from the Roessler tables (Roessler et al., 1978) to determine
a detection threshold for 4-ethylcatechol in white and red wine, and estimated this threshold in
the range of 100 – 400 µg/L. However, their research does not specifically couple a detection
42
threshold to either red or white wine. It is interesting to note the vast differences between the
three threshold values available for 4-ethylcatechol: 60 µg/L (Hesford & Schneider, 2004), 100 –
400 µg/L (Larcher et al., 2008) and 774 µg/L (Curtin et al., 2008). These discrepancies warrant
further investigation into this compound’s effect in wine.
Ferreira et al. (2000) determined the detection thresholds for isobutyric acid, isovaleric
acid and 4-ethylguaiacol in a synthetic wine solution using a mixed triangle test. The synthetic
wine solution contained 11% v/v ethanol, 7 g/L glycerin, 5 g/L tartaric acid and a pH of 3.4. The
level at which 50% of the panellists could recognise the difference for the control were taken as
the detection threshold value. The thresholds determined were 2300 µg/L, 33.4 µg/L and 33
µg/L for isobutyric acid, isovaleric acid and 4-ethylguaiacol respectively. However, in terms of
the validity of this test, a similar argument as that applied to the research of Chatonnet et al.
(1992) holds true.
Finally, the detection thresholds for acetic acid and 4-vinylguaiacol were determined by
Guth (1997) in a water and ethanol mixture where the ratio was 9:1, through a method that is
only described as “nasal comparison”. A detection threshold that of was 2 000 000 µg/L, or 0.2
g/L was found for acetic acid. The result found for 4-vinylguaiacol was 40 µg/L. These results
are questionable, as no accepted sensory or statistical method was applied.
The American Standard Test Manual (ASTM) E679 method has been used to determine
detection thresholds by several other authors, both in wine and in other media. Examples in
wine include the determination of the detection threshold for rotundone (Wood et al., 2008), oak
lactones (Brown et al., 2006) and diacetyl (Martineau et al., 1995). This method involves
presenting judges with 3-alternative forced choice tests in increasing concentration until the odd
sample has been correctly identified in two consecutive cases. This gives rise to a best estimate
threshold (BET) value, which is the geometric mean of the last incorrectly identified sample and
the first correctly identified sample. The test is performed over a set range of concentrations,
which is determined beforehand by means of sensory testing. However, as the test evaluates a
predetermined concentration range, the data produced are both top and bottom truncated. This
means that the range in which the test functions has both a top and a bottom limit, which can be
problematic if the detection threshold falls close to these limits.
The determination of odour thresholds uses the combined sensory response of a
selected group of trained individuals, called panellists. However, the correct identification of an
odorant at very low concentration levels in a specific wine can pose several challenges. These
include the varying sensitivity of panel members (a factor affected by physiological differences
or professional experience), tiredness of sense organs, temporal persistence of a characteristic
aroma, as well as synergistic effects of different compounds on the sensory character of a wine
(Le Berre et al., 2007). Furthermore, specific compounds such as alcohol or other aromatic wine
compounds – especially fruity and woody odourants – can have a significant effect on the
perception of Brett character in different media (Lawless, 1999; Le Berre et al., 2007; Escudero
43
et al., 2007). This has been demonstrated for Brett character by the research of Curtin et al.
(2008). However, it has been shown that the use of training has a positive effect on the
accuracy of data determined from triangle tests (Dacremont & Sauvageot, 1997).
The aim of this study was to determine the detection thresholds in Pinotage red wine for
eight compounds, namely 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, isovaleric acid,
isobutyric acid, 4-vinylphenol, 4-vinylguaiacol and acetic acid. This served several purposes.
Firstly, none of the detection thresholds of these compounds have been determined in Pinotage
red wine to date. Pinotage, which was bred in South Africa, forms an important part of South
African red wine’s portfolio and the determination of these values thus has a potential benefit for
the South African wine industry. Furthermore, as described earlier, for some of these
compounds, the only literature values available are for detection in either water or model
solution (Guth, 1997; Ferreira et al., 2000; Curtin et al., 2005), and it has been shown that
detection thresholds found in model solutions are not truly comparable to those determined in
wine (Le Berre et al., 2007). Wine style and type has also generally been found to have an
effect on detection threshold (Martineau et al., 1995; Brown et al., 2006). Finally, as the sensory
effects of these compounds were tested in greater detail in subsequent studies (see Chapter 4);
it was of utmost importance to initially determine the detection thresholds of these compounds in
Pinotage.
2 MATERIALS AND METHODS
2.1 18BSamples
Three hundred litres of Pinotage red wine was supplied by a local producer of wines (Distell
Group Ltd, Stellenbosch, South Africa) during the course of 2008 and was bottled manually at
the Department of Viticulture and Oenology, Stellenbosch University, South Africa. The wine
was made using standard red wine making practices and completely underwent both alcoholic
and malolactic fermentation. The wine had a pH of 3.7, and an alcohol concentration of 12.9 %.
The wine had not been wooded prior to bottling, and had an aroma profile that was dominated
by fruit.
After bottling, samples were taken to determine the levels of the compounds investigated
in this study in the wine. Analysis of 4-ethylphenol, 4-ethylguaiacol, 4-vinylphenol, 4-
vinylguaiacol, isovaleric acid, isobutyric acid and acetic acid were performed using GC-MS. The
analysis for 4-ethylcatechol was performed using HPLC-MS. All these analyses were conducted
by an accredited wine analysis laboratory (Quantum Laboratories, South Africa).
The wine used in this study was found to contain 6 µg/L 4-ethylphenol, 4 µg/L 4-
ethylguaiacol and 0 µg/L 4-ethylcatechol. The wine was therefore considered to be free of
44
Brettanomyces spoilage. However, the wine contained 370 µg/L 4-vinylphenol, and 29 µg/L 4-
vinylguaiacol. Both these values are rather high. The isovaleric acid concentration was 168 µg/L
and the isobutyric acid concentration 68 µg/L. Both these values are well below those typically
found in red wine (Francis & Newton, 2005). Finally, the acetic acid concentration was 0.20 g/L.
Table 3.1. Concentration ranges used for spiking wines during the determination of detection
thresholds of Brettanomyces related compounds in Pinotage wine.
Concentration used Compound MFa
1 2 3 4 5 6 7 8
4-ethylphenol (μ/L) 1.4 98 137.2 192 268 376 527 737 1033
4-ethylguaiacol (μ/L) 1.4 71 99 139 195 273 382 535 748
4-ethylcatechol (μ/L) 1.4 99 139 194 272 380 532 745 1044
Isovaleric acid (μ/L) 1.7 12 20.4 34 58 100 170 289 492
Isobutyric acid (μ/L) 1.7 260 442 751 1277 2171 3691 6275 10668
4-vinylphenol (μ/L) 1.25 6.4 8 10 12.5 15.6 19.5 24.4 30.5
4-vinylguaiacol (μ/L) 1.3 5 6.5 8.45 10.9 14.28 18.56 24.1 31.3
Acetic acid (g/L) 1.5 0.1 0.15 0.225 0.337 0.506 0.759 1.139 1.708 aMultiplication factor: factor used for multiplication between different levels
Solutions of 10 mg/mL, 1 mg/mL, 100 µg/mL and 10 µg/mL of the eight compounds (4-
ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, 4-vinylphenol, 4-vinylguaiacol, isovaleric acid,
isobutyric acid and acetic acid) (Aldrich, South Africa) were prepared in 99.5% ethanol (Merck
Chemicals, South Africa). These solutions were used to “spike” the wines with the
concentrations listed in XTable 3.1X so that the final concentrations in the wines correlated to
those listed in XTable 3.2 X. Concentration ranges for all eight of the compounds involved were
sourced from literature, and were originally evaluated by a panel of three to four wine industry
consultants (Thales South Africa, IWBT, Stellenbosch University) who are considered experts in
the field of “Brett character.” These concentration ranges consisted of 11 concentrations, with
the literature maximum set at approximately level 8, and the literature detection threshold set at
approximately level 5. These samples were analysed by means of sensory analysis by
consensus using the panel of consultants. The final concentrations used in this study were
decided upon during these analyses, and are shown in XTable 3.2 X.
2.2 19BDetermination of detection threshold levels
The determination of the detection threshold levels was carried using a method based on the
standard method of the American Standard Test Manual (ASTM E 679 – 04). This method is the
standard practice for the determination of odour and taste thresholds by a forced choice
ascending concentration series method of limits. The test is defined as a three-alternative forced
45
choice (3-AFC) test, and involves presenting judges with sets of three samples, of which one
contains the compound in question. Judges are then instructed to identify the odd sample in
each set. Sample sets are presented in ascending order of concentration. This ASTM method
prescribes that the test should be terminated as soon as a judge made two consecutive correct
choices, in other words, as soon as the “spiked” sample was correctly identified in two
consecutive concentrations. A more detailed description of the test, the training involved, the
modifications to the test and the data analysis method used follow in the appropriate sections.
Table 3.2. Final concentration ranges used in the determination of detection thresholds of
Brettanomyces related compounds in Pinotage wine.
Concentration used Compound MFa
1 2 3 4 5 6 7 8
4-ethylphenol (μ/L) 1.4 104 143.2 198 274 382 533 743 1039
4-ethylguaiacol (μ/L) 1.4 75 103 143 199 277 386 539 752
4-ethylcatechol (μ/L) 1.4 99 139 194 272 380 532 745 1044
Isovaleric acid (μ/L) 1.7 180 188.4 202 226 268 338 457 660
Isobutyric acid (μ/L) 1.7 328 510 819 1345 2239 3759 6343 10736
4-vinylphenol (μ/L) 1.25 376 378 380 382.5 385.6 389.5 394.4 400.5
4-vinylguaiacol (μ/L) 1.3 34 35.5 37.45 39.9 43.28 47.56 53.1 60.3
Acetic acid (g/L) 1.5 0.3 0.35 0.425 0.537 0.706 0.959 1.339 1.908 aMultiplication factor: factor used for multiplication between different levels
2.2.1 47BSubjects and training
A trained panel was used for the determination of detection thresholds of the eight compounds
(4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, 4-vinylphenol, isovaleric acid, isobutyric acid
and acetic acid) in the Pinotage red wine described above. The panel consisted of 10
individuals who had experience participating in sensory tests. A portion of the panel had
previously participated in an extensive study for the determination of the detection thresholds of
cork-taint related compounds, and therefore was experienced in sensory analysis of wine, as
well as the use of the triangle test. However, at the onset of the project the panel had no or
limited experience in detecting Brett character in wine and the ten judges were thus trained
extensively in the detection of the above-mentioned compounds in Pinotage. The ASTM test is
a 3-AFC test, which requires that the differences between samples should be known prior to the
analysis (O’Mahony, 1995). Training of judges was therefore motivated by the fact that the
aromas associated with the different compounds are extremely difficult to communicate verbally.
The panel was trained in two phases. In Phase 1 each judge received a control sample
containing only the base wine, as well as three wine samples and one water sample spiked with
the compound in question. The first wine sample contained the lowest level that were be used in
46
the test (level 1, see XTable 3.1 X). The second wine sample contained the compound in question
at the level closest to its assumed detection threshold (usually level 4 or 5, XTable 3.1 X). The third
wine sample contained the compound in question at the highest level that will be tested (level 8,
see XTable 3.1 X). The water sample, which was used as a reference sample, contained the
compound at the highest level that will be tested (level 8, see XTable 3.1 X). These samples were
used to characterise the aromas of the specific compounds, and to familiarise the judges with
the particular aroma associated with each compound. The judges received the samples in a
“round-table” situation where the differences between the wines were discussed and descriptors
identifying the differences between the samples were generated. The identification of difference,
as well as the generation of descriptors, was an essential part of performing the test, as the
difference between samples should be known for the performance of a true 3-AFC test
(O’Mahony, 1995).
In Phase 2 of training each judge received eight sets of samples, each set containing
three samples. In each set, two of the three samples contained only the base wine (untainted
wine) and a third sample in every set contained the base wine plus the added compound. The
concentration increased with a constant factor from set 1 through to set 8 as illustrated in XTable
3.1 X. The sets were presented in an order of ascending concentration and the samples within
each set were presented in a randomised order. The sample volume was 20 mL and all the
samples were served in ISO wine tasting glasses at 20 ±1°C. Each sample was numbered with
a random three-digit code and was covered with a tight-fitting lid to prevent the aroma from
escaping and contaminating the laboratory environment.
The judges were instructed to smell the headspace of the samples in each set, i.e. in the
order presented and then to indicate the odd or tainted sample. Each judge was required to rest
for a total of 2 min between every set and 5 min between the fourth and fifth set which was
regarded as the half-way mark for this procedure. The latter was to cancel strong carry-over
effects and to minimise tiredness of the sense organs. This procedure was repeated until
consensus was reached on the odd sample within each set. Phase 2 of training was performed
for each of the eight compounds used in this study.
2.2.2 48BDetermination of detection thresholds
The detection thresholds of the different compounds were determined by the panel as described
in Phase 2 of the training. The panel determined the detection levels of all eight compounds (4-
ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, 4-vinylphenol, isovaleric acid, isobutyric acid and
acetic acid) in Pinotage red wine. The final analyses were conducted by 10 trained assessors in
booths with standard artificial daylight lighting and temperature control at 20°C ±1°C.
Communication was not allowed between the judges for the duration of the test. Although the
ASTM stipulates terminating the test after a judge detected a compound, the judges received all
47
eight sample sets, and their responses were only assessed after the test, as the responses of
the judges could not be assessed during the test. Each detection threshold test consisted of
four replicates of the ASTM E679 - 04 test.
2.2.3 49BAnalysis of data
An example of the results obtained from this test is shown in XTable 3.3X. The ASTM E679–04
method prescribes that “detection” is defined as correctly identifying the spiked sample from the
two unspiked samples in two consecutive sample sets. “0” indicates incorrect identification of
the spiked sample, and “+” indicates correct identification of the spiked sample. The levels
marked with the superscript “a”, are the levels at which the judges detected the compound
according to the ASTM method. All data regarding further sample sets are ignored. From these
results, the detection thresholds are calculated in two steps.
Table 3.3. Example of results obtained from ASTM method. Note that the concentrations are
hypothetical. “0” indicates incorrect identification of the spiked sample, and “+” indicates correct
identification of the spiked sample.
Level Judge
1 2 3 4 5 6 7 8 BETb, c
1 0 0 0 0a + + + + (c4 x c5)1/2
2 0 0 + 0 0a + + + (c5 x c6)1/2
3 0 + 0a + + + + + (c3 x c4)1/2
4 0 0 0a + + + + + (c3 x c4)1/2
5 0 0a + + 0 + + + (c2 x c3)1/2
6 + + + + 0 0 0 0 (c0 x c1)1/2
7 0 + 0 0 + 0 + 0a (c8 x c9)1/2
8 0a + + 0 0 + + + (c1 x c2)1/2
a Level which is considered the first instance of “detection”. In the case of Judge 6, the level that would have preceded level 1 is considered to be the first instance of “detection”. b BET = Best Estimate Threshold value obtained in specific instance of the test. c c designates the value of the actual concentration used in the test for that specific level.
Firstly, the best estimate threshold (BET) is determined per judge per sample set. This is
calculated using equation (1).
1)
Where:
BET = Best estimate threshold
c1 = last missed concentration
c2 = next concentration after c1
21 ccBET
48
Simply put, the BET values are the geometric mean of the last concentration that the
judge could not detect, and the first concentration that was considered as “detected”. The
geometric mean is used in order to compensate for the fact that the concentration values used
fall on a logarithmic scale.
Once the BET values for all judges and replications were calculated, the detection
thresholds were calculated either using the median or the prescribed ASTM E679–04 method.
The calculation used for the ASTM E679–04 method is show in equation (2).
2)
Where:
DT = detection threshold
n = number of repetitions (Judge x Replications)
In other words, the detection threshold is calculated as the geometric mean of the BET
values over all judges and replicates. A simple method of calculating this is by finding the mean
of the logarithms of the BET values, and then finding the antilog of the mean. When the median
was used, the median of all the logarithmic values was found, and the antilog of these values
was said to be the detection threshold.
A 95% confidence interval was calculated for both the ASTM method and for the
median. The confidence interval of a value is an interval in which it can be said that the estimate
of the value falls with a quantitative degree of statistical certainty. The range of the interval can
be considered as a measure of accuracy for a statistical method: a method with a smaller range
is considered to be more accurate (Neyman, 1937). The confidence interval of the median was
calculated using the method described by (Snedecor & Cochran, 1967) by finding the order
statistics with the following numbers:
3)
22
1 nznULos
4)
22
1 nznLLos
Where:
ULos = Order statistics for upper confidence limit
LLos = Order statistics for lower confidence limit
z = normal deviate corresponding to the desired probability (in this case 0.05)
n
n
BETDT 1
49
n = number of responses
Note that the order statistic does not refer to the actual value of the confidence limit, but
to the ranking of a number that corresponds to this confidence limit.
3 RESULTS AND DISCUSSION
3.1 20BComparison between different calculation methods
The results obtained from the two different calculation methods are compared in XTable 3.4X. Note
that base wine concentrations are not taken into account in these results. These results are
shown visually in Figures 3.1 to 3.8.
Table 3.4. Comparison of two calculation methods for determination of detection thresholds. In
this table, LL designates the lower confidence limit of each value, and UL designates the upper
confidence limit. Note that base wine concentrations are not taken into account in these results.
Compound ASTM
LL
ASTM Value
ASTM
UL
Median LL
Median Value
Median UL
4-ethylphenol (μ/L) 174 221 281 162 195 250
4-ethylguiacol (μ/L) 87 107 130 60 84 117
4-ethylcatechol (μ/L) 316 442 618 229 385 881
Isovaleric acid (μ/L) 44 72 117 26 44 222
Isobutyric acid (μ/L) 1132 1756 2726 576 1666 4813
4-vinylphenol (μ/L) 10 14 20 10 11 26
4-vinylguaiacol (μ/L) 9 12 16 7 11 28
Acetic acid (g/L) 0.165 0.204 0.252 0.122 0.108 0.298
Some general trends can be observed from all these datasets. Firstly, the median gives
a lower estimation of detection threshold in all cases, but in many cases the estimation given by
the median falls between the same two levels than the estimation given by the ASTM method.
In some cases, (for example acetic acid, XFigure 3.8 X) the difference between the median and its
lower level is smaller than the difference between the ASTM and its lower level. However, in
almost all of the datasets, the difference between the median and its upper limit (up to 5 levels)
is significantly larger than the difference between the ASTM and its upper limit (usually 1 level).
Although the use of the median generally gives a larger confidence interval than the
ASTM method and may therefore be considered to be less accurate (Neyman, 1937), the
median and its confidence limits give a much better snapshot of the performance or abilities of
the sensory panel, whereas the ASTM method may be an oversimplified method of calculating
detection thresholds.
50
LLLL
ASTM
Median
UL
UL
0
50
100
150
200
250
300
350
400
Median ASTM
Calculation method
Co
nce
ntr
atio
n (μ
g/L
)Level 5
Level 2
Level 4
Level 3
Level 1
Figure 3.1. Effect of calculation methods on detection threshold of 4-ethylphenol. LL designates
the lower confidence limit of each value, and UL the upper confidence limit.
LL
LL
ASTM
Median
UL
UL
0
50
100
150
200
Median ASTM
Calculation method
Co
nce
ntr
atio
n (μ
g/L
)
Level 1
Level 2
Level 3
Level 4
Figure 3.2. Effect of calculation method on detection threshold of 4-ethylguaiacol. LL
designates the lower confidence limit of each value, and UL the upper confidence limit.
51
LL
LL
ASTMMedian
UL
UL
0
200
400
600
800
1000
1200
Median ASTM
Calculation method
Co
nce
ntr
atio
n (μ
g/L
)
Level 3
Level 4
Level 5
Level 6
Level 7
Level 8
Level 2Level 1
Figure 3.3. Effect of calculation method on detection threshold of 4-ethylcatechol. LL
designates the lower confidence limit of each value, and UL the upper confidence limit.
Figure 3.4. Effect of calculation method on detection threshold of isovaleric acid. LL designates
the lower confidence limit of each value, and UL the upper confidence limit.
Level 3
Level 1
LLMedian
ASTM
UL
UL
0
50
100
150
200
250
300
Median ASTM
Calculation method
Level 2
Level 4
Level 5
Level 6
Level 7
LL
Con
cent
ratio
n (μ
g/L)
52
LL LL
ASTM
Median
UL
UL
0
5
10
15
20
25
30
35
Median ASTM
Calculation method
Co
nce
ntr
atio
n (
ug
/L)
Level 1
Level 2
Level 3
Level 4
Level 5
Level 6
Level 7
Level 8
Figure 3.5. Effect of calculation method on detection threshold of 4-vinylphenol. LL designates
the lower confidence limit of each value, and UL the upper confidence limit.
LL
LL
ASTMMedian
UL
UL
0
5
10
15
20
25
30
35
Median ASTM
Calculation method
Co
nce
ntr
atio
n (
ug
/L)
Level 2
Level 3
Level 4
Level 5
Level 6
Level 7
Level 8
Level 1
Figure 3.6. Effect of calculation method on detection threshold of 4-vinylguaiacol. LL designates
the lower confidence limit of each value, and UL the upper confidence limit.
53
LL
LL
Median ASTM
UL
UL
0
1000
2000
3000
4000
5000
6000
7000
Median ASTM
Calculation method
Co
nce
ntr
atio
n(u
g/L
)
Level 2Level 3
Level 4
Level 5
Level 6
Level 7
Level 1
Figure 3.7. Effect of calculation method on detection threshold of isobutyric acid. LL designates
the lower confidence limit of each value, and UL the upper confidence limit.
LL
LL
ASTM
Median
UL
UL
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Median ASTM
Calculation method
Co
nce
ntr
atio
n (
g/L
)
Level 1
Level 2
Level 3
Level 4
Figure 3.8. Effect of calculation method on the detection threshold of acetic acid. LL designates
the lower confidence limit of each value, and UL the upper confidence limit.
54
The median gives an indication where most of the instance of detection occurred,
whereas the ASTM method produces an “average” result. This is particularly important as the
data being analysed is both top and bottom truncated and can therefore only fall into a
predetermined range. This means that detecting a compound at the lowest level may indicate
that the individual could have detected this compound below the lowest level presented to
judges. However, this effect is “averaged out” when the ASTM method is used.
The median also gives a good indication of the general ability of each compound to be
detected by individuals. A median with an upper limit that is much higher than the median
indicates that although most individuals detected the compound at a certain level, it could only
be detected at higher concentrations in several other instances. This gives an indication of the
distribution of the variability in detection ability of a specific compound between judges. For
example, the lower detection limit and the detection limit of 4-vinylphenol (XFigure 3.5X) both lie
between level 3 and 4, whereas the upper detection limit lies between level 7 and 8. This
indicates that most individuals were able to detect this compound at between level 3 and 4;
some individuals were only able to detect it at much higher levels.
The differences in results obtained by these two calculation methods raises the question
of the true definition of the term “detection threshold”. In ASTM E679-04, detection threshold is
defined as “the lowest concentration of a substance in a medium relating to the lowest physical
intensity at which a stimulus is detected as determined by the best-estimate criterion”. This
implies that the detection threshold of a compound should be set at the lowest level that the
compound can be detected by most individuals, regardless of whether or not other individuals
are less able to detect this compound at those concentrations. If this is taken into account, the
median appears to be a more suitable method for the calculation of detection threshold, as the
result obtained from this calculation method is not as affected by the variation between judges.
An alternative method of compensating for the varied abilities of judges may lie in
identifying judges that perform poorly or inconsistently, and subsequently omitting their
observations from the data set. Looking at the dataset as a whole may give a much better
indication of whether or not a judge truly detected the compound or if the two consecutive
correct indications required for “detection” occurred simply by chance. It is recommended that
future studies focus on the diagnostics of detection-threshold type data and that a diagnostic
test be developed for sets of detection threshold data. Such a test should not only take into
consideration the probability correctly identifying the “spiked” sample by chance, but also the
probability of it occurring twice, three times and so forth, and the statistical implications of
incorrect indications once two consecutive correct identifications have occurred.
A further statistical issue that could not be addressed in this study but deserves future
attention is the type of detection threshold test used. The ASTM E679-04 is defined as using a
three-alternative forced choice (3-AFC) test, which it loosely describes as “a set consisting of
one test sample and two blank samples”. However, the definition of the 3-AFC test is not quite
55
as simple as the standard makes it out to be. O’Mahony (1995) defines a 3-AFC test as a test
where the sensory differences between the samples are known to the judges performing the
test. He also states that the probability of a judge correctly identifying an odd sample in a 3-AFC
test is significantly higher than in a simple triangle test where the difference is not known. This
difference in performance and probability is of a sufficient magnitude to give different outcomes
should binomial proportional statistics like the tables developed by Roessler et al., (1978) be
applied. This means that in some cases, the application of a true 3-AFC test would have the
outcome that the samples are significantly different according to the Roessler tables, but
applying a normal triangle test (where the difference is not known) would have the outcome that
the samples are not significantly different.
The implications for this study are that although the principles for performing a 3-AFC
test were followed, severe difficulties were experienced in defining the difference in sensory
profile caused by some of the compounds. Two specific examples were 4-ethylcatechol and
isobutyric acid as for both these compounds a difference in the aroma profile of the wine could
be perceived, but the difference was not easily identifiable. In the case of other compounds, like
4-ethylguaiacol, the difference in sensory profile was immediately identified by all the judges.
The implication of this is that the statistical principles on which ASTM E679-04 are based may
not be equally appropriate for all instances that this test may be used.
3.2 21BComparison to literature
The results obtained with both calculation methods are compared with their literature values in
XTable 3.5 X. The “total” results (the detection thresholds obtained plus the concentration present
in the base wine) are shown in XTable 3.6 X.
As can be seen in XTable 3.5 X, the detection thresholds found using the median method
during this study are generally lower than those reported in literature. However, when the levels
of these compounds present in the wine prior to addition is taken into consideration (XTable 3.6 X),
detection threshold are lower in some cases (4-ethylphenol and isobutyric acid), comparable in
some cases (4-vinylguaiacol and 4-ethylguaiacol) and higher in other cases.
The values obtained for 4-vinylguaiacol and 4-ethylguaiacol (XTable 3.6 X) are similar to
those obtained from literature (Chatonnet et al., 1992; Guth, 1997). However, the median
obtained for 4-ethylguaiacol is significantly lower than that obtained through the ASTM. From
this it could be concluded that there existed a difference in sensitivity between judges in terms
of the perception of 4-ethylguaiacol. This is in line with the findings of Laska & Hudson (1991)
and Curtin et al. (2008).
56
Table 3.5. Comparison of detection thresholds obtained with two calculation methods to
literature.
Compound Median ASTM Literature
value Reference
4-ethylphenol (μg/L) 195 221 Range: 162 - 250 174 - 281
605a Chatonnet et al., 1993
4-ethylguaiacol (μg/L) 84 107 Range: 60 - 117 87 - 130
110a Chatonnet et al., 1993
4-ethylcatechol (μg/L) 385 442 Range: 229 - 881 316 - 618
60a Hesford & Schneider, 2004
Isovaleric acid (μg/L) 44 72 Range: 26 - 222 44 - 117
33b Ferreira et al., 2000
Isobutyric acid (μg/L) 1666 1756 Range: 676 - 4813 1132 - 2726
2300b Ferreira et al., 2000
4-vinylphenol (μg/L) 11 14 Range: 10 - 26 10 - 20
180b Culleré et al., 2003
4-vinylguaiacol (μg/L) 11 12 Range: 7 - 28 9 - 16
40b Guth, 1997
Acetic acid (g/L) 0.122 0.204 Range: 0.108 – 0.207 0.165 - 0.252
0.20b Guth, 1997
a Determined in red wine. b Determined in model solution.
The total detection thresholds (XTable 3.6X) of isovaleric acid determined using both
methods were both significantly higher than the detection threshold obtained from literature
(Ferreira et al., 2000). This is a direct result of the fact that the wine used in this study contained
a fair amount of this compound. The higher detection threshold can also be ascribed to the fact
that the literature detection threshold was determined in model solution, and that it is generally
more difficult to detect odorants in real wines than in model solution (Le Berre et al., 2007). The
higher detection thresholds for 4-vinylphenol and acetic acid found in this study compared to
literature (Guth, 1997; Culleré et al., 2003) are further evidence of this phenomenon.
Both detection thresholds ( XTable 3.6 X) determined for 4-ethylphenol were substantially
lower than that reported in literature. This can be ascribed to the fact that the wine we used was
not strongly wooded, and 4-ethylphenol falls in the same semantic category as woody
odourants (Escudero et al., 2007). This means that for wooded wines, the wood character
suppresses some of the aroma character of 4-ethylphenol, making it more difficult to detect.
This is in agreement with the findings of Curtin et al. (2008). Conversely, 4-ethylphenol is easier
to detect in a wine with a lower level of woodiness, leading to a lower detection threshold in this
study.
Another relevant sensory aspect of the wine used is the fact that it was a very fruity
wine, which may also affect the detection thresholds of certain compounds. Isobutyric acid was
described by the panel as having a fruity/floral character, and it is possible that the lower
detection threshold found for isobutyric acid is a result of the amplifying effect that it may have
on the already high fruitiness of the wine, making it easier to detect at lower concentrations.
57
Guadagni et al. (1963) postulated a theory of additive effects between odour compounds, which
stated that detection thresholds of compounds are lower when they are present in solution with
other compounds. Although it has been found (Grosch, 2001) that this theory does not hold true
in all cases, synergism has been found between specific odour pairs (Laing, 1988; Laska &
Hudson, 1991). The effect exhibited by isobutyric acid in this study may be an example of such
a type of synergism.
Table 3.6. Total detection thresholds (detection thresholds obtained plus base wine
concentration).
Compound Median ASTM Literature
value Reference
4-ethylphenol (μg/L) 201 221 Range: 168 - 256 180 – 287
605a Chatonnet et al., 1993
4-ethylguaiacol (μg/L) 84 111 Range: 64 - 121 91 - 134
110a Chatonnet et al., 1993
4-ethylcatechol (μg/L) 385 442 Range: 229 - 881 316 - 618
60a Hesford & Schneider, 2004
Isovaleric acid (μg/L) 214 242 Range: 196 - 392 214 - 287
33b Ferreira et al., 2000
Isobutyric acid (μg/L) 1735 1825 Range: 676 - 4813 1132 - 2726
2300b Ferreira et al., 2000
4-vinylphenol (μg/L) 381 384 Range: 380 - 396 380 - 390
180b Culleré et al., 2003
4-vinylguaiacol (μg/L) 40 41 Range: 36 - 57 38 - 45
40b Guth, 1997
Acetic acid (g/L) 0.322 0.404 Range: 0.308 – 0.407 0.365 - 0.452
0.20b Guth, 1997
a Determined in red wine. b Determined in model solution.
The highest discrepancy between threshold values obtained here and reported in
literature was observed for 4-ethylcatechol. This may be due to the fact that Hesford and
Schneider (2004) did not use an accepted methodology for the determination of the detection
threshold of 4-ethylcatechol, as their article gave no details for the method that they employed
for the determination of this detection threshold. A more recent study, by Larcher et al. (2008)
also struggled to detect 4-ethylcatechol at the concentration found by Hesford and Schneider
(2004), and placed the detection threshold for 4-ethylcatechol in the range of 100 – 400 µg/L,
which is more in line with our observations. However, the value that this research found was
much lower than the value of 774 µg/L found by Curtin et al. (2008). This may be due to the test
method used. As Curtin et al. (2008) used Cabernet Sauvignon wine in their study, which could
be less prone to being affected by 4-ethylcatechol than a fruity, unwooded Pinotage.
4 CONCLUSIONS
58
In this study, the detection thresholds for eight different compounds associated with
Brettanomyces were determined. Two data analysis methods were compared. It was found that
although using the median gives a lower detection threshold and a larger confidence interval, it
gives a better indication of overall panel performance. Some discrepancies were found between
the literature values and those determined experimentally. Most of these discrepancies could be
clarified by an investigation into the methods employed to establish these threshold values, as
well as the aroma profiles of the wine used. These discrepancies, as well as the reasons for the
differences, justified the determination of these detection thresholds before proceeding to
further sensory studies.
5 REFERENCES
ASTM (1997). Standard practice E 679-91. Determination of odor and taste thresholds by an
ascending concentration series method of limits. Philadelphia, USA: American Society
for Testing and Materials.
Brown, R. C., Sefton, M. A., Taylor, M. A., Taylor, D. K. & Elsey, G. M (2006) An odour
detection threshold determination of all four possible stereoisomers of oak lactone in a
white and red wine. Australian Journal of Grape and Wine Research. 12, 115 - 118.
Chatonnet, P., Dubourdieu, D. & Boidron, J. (1995) The influence of Brettanomyces/Dekkera sp.
yeasts and lactic acid bacteria on the ethylphenol content of red wines. American
Journal of Enology and Viticulture. 46, 463 - 468.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Lavigne, V. (1993) Synthesis of volatile phenols by
Saccharomyces cerivisae in wines. Journal of the Science of Food and Agriculture. 62,
191 - 202.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
Culleré, L., Escudero, A., Cacho, J., Ferreira, V. (2004) Gas chromatography-ilfactometry and
chemical quantitative study of the aroma of six premium quality Spanish aged red wines.
Journal of Agricultural and Food Chemistry. 52, 1653 - 1660.
Curtin, C. Bramley, B. Cowey, G. Holdstock, M. Kennedy, E. Lattey, K. Coulter, A. Henschke, P.
Francis, L. Godden, P. Sensory perceptions of 'Brett' and relationship to consumer
preference. Blair, R.J.; Williams, P.J.; Pretorius, I.S. Proceedings of the thirteenth
Australian wine industry technical conference, 29 July-2 August 2007, Adelaide, SA. :
207-211; 2008
Curtin, C. D., Bellon, J. R., Coulter, A. D., Cowey, G. D., Robinson, E. M. C., de Barros Lopes,
M. A., Godden, P. W., Henschke, P. A. & Pretorius, I. S. (2005) The six tribes of "Brett"
in Australia - Distribution of genetically divergent Dekkera bruxellensis strains across
59
Australian winemaking regions. Australian and new Zealand Wine industry Journal. 20,
28 - 35.
Dacremont, C. & Sauvageot, F. (1997) Are replicate evaluations of triangle tests during a
session good practice? Food Quality and Preference. 8, 367 - 372.
du Toit, W. J., Pretorius, I. S. & Lonvaud-Funel, A. (2005) The effect of sulphur dioxide and
oxygen on the viability and culturabilty of a strain of Acetobacter pasteurianus and a
strain of Brettanomyces bruxellensis isolated from wine. Journal of Applied Microbiology.
98, 862 – 871.
Escudero, A., Campo, E., Fariña, L., Cacho, J. & Ferreira, V. (2007) Analytical characterization
of the aroma of five premium red wines. Insights into the role of odor families as the
concept of fruitiness of wines. Journal of Agricultural and Food Chemistry. 55, 4501 -
4510.
Fariña, L., Boido, E., Carrau, F. & Dellacassa, E. (2007) Determination of volatile phenols in red
wines by dispersive liquid-liquid microextraction and gas chromatography-mass
spectrometry detection. Journal of Chromatography A. 1157, 46 - 50.
Ferreira, V., Lόpez, R. & Cacho, J. F. (2000) Quantitative determination of the odorants of
young red wines from different grape varieties. Journal of the Science of Food and
Agriculture. 80, 1659 - 1667.
Francis, I. L. & Newton, J. L (2005) Determining wine aroma from compositional data. In: AWRI
– Advances in wine science (edited by R. J. Blair, M. E. Francis & I. S. Pretorius A. L.),
Pp 201 - 212. Glen Osmond, Australia: Australian Wine Research Institute.
Fugelsang, K. C. & Zoecklein, B. W. (2003) Population dynamics and effects of Brettanomyces
bruxellensis strains on Pinot noir (Vitis vinifera L.) wines. American Journal of Enology
and Viticulture. 54, 294 - 300.
Grosh, W. (2001) Evaluation of the key odorants of foods by dilution experiments, aroma
models and omission. Chemical Senses. 26, 533 - 545.
Guadagni, D. G., Buttery, R. G., Okano, S. & Burr, H. K. (1963) Additive effect of sub-threshold
concentrations of some organic compounds associated with food aromas. Nature. 200,
1288-1289.
Guth, H (1997) Quantification and sensory studies of character impact odorants of different
white wine varieties. Journal of Agricultural and Food Chemistry. 45, 3027-3032.
Henschke, P., Bellon, J., Capone, D., Coulter, A., Cowey, G., Cozzolino, D., Curtin, C., Field, J.,
Gishen, M., Graves, P., Latey, K., Robinson, E., Francis, I. L., Lopes, M, D. & Godden,
P. (2004) Incidence and control of Brettanomyces, the Australian perspective. American
Journal of Enology and Viticulture. 55, 304A.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11-13.
60
Hesford, F., Schneider, K., Porret, N. & Gafner, J. (2004) Identification and analysis of 4-ethyl
catechol in wines tainted by Brettanomyces off-flavor. American Journal of Enology and
Viticulture. 55, 304A.
Laing, D. G. (1988) Relationship between the differential adsorption of odorants by the olfactory
mucus and their perception in mixtures. Chemical Senses. 13, 463 - 471.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determinination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Laska, M. & Hudson, R. (1991) A comparison of the detection thresholds of odour mixtures and
their components. Chemical Senses. 16, 651 - 662.
Lawless, H. T. & Heymann, H. (1998). Sensory evaluation food, principles and practice. New
York, USA: Chapman and Hall.
Lawless, H. T. (1999) Descriptive analysis of complex odors: reality, model or illusion. Food
Quality and Preference. 10, 325 - 332.
Le Berre, E., Atanasova, B., Langloi, Etiéviant, P. & Thamas-Danguin, T. (2007) Impact of
ethanol on the perception of wine odorant mixtures. Food Quality and Preference. 18,
901 - 908.
Licker, J. L, Acree, T. E. & Henick-Kling, T. (1999) What is "Brett" (Brettanomyces) flavour? A
preliminary investigation. In: Chemistry of Wine Flavour. ACS Symposium Series (edited
by A. L. Waterhouse & S. E. Ebeler), Pp 96 - 115. Washington DC: American Chemical
Society.
Loureiro, V. & Malfeito-Ferreira, M. (2003) Spoilage yeasts in the wine industry. International
Journal of Food Microbiology. 86, 25 - 20.
Martineau, B., Acree, T. E. & Henick-Kling, T. (1995) Effect of wine type on detection threshold
for diacetyl. Food Research International. 28, 139 - 143.
Neyman, J. (1937) Outline of a theory of statistical estimation based on the classical theory of
probability. Philosophical Transactions of the Royal Society of London. Series A,
Mathematical and Physical Sciences. 236, 333 - 380.
O'Mahony, M. (1995) Who told you the triangle test was simple? Food Quality and Preference.
6, 227 - 738.
Rayne, S., & Eggers, N. (2007) Volatile phenols from Brettanomyces in barreled red wines: are
they stable? And where do they reside? The Australian & New Zealand Grapegrower &
Winemaker. 522, 64 - 69.
Roessler, E. B., Pangborn, R. M., Sidel, J. L. & Stone, H. (1978) Expanded statistical tables for
estimating significance in paired-preference, paired-difference, duo-trio and triangle
tests. Journal of Food Science. 43, 940 - 947.
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of “Brett character”. Food Chemistry. 114, 15 - 19.
61
Snedecor, G. W. & Cochran, W.G. (1967) Shortcut and non-parametric methods. In: Statistical
Methods, Pp 121 – 134. Ames, Iowa, USA: The Iowa State University Press.
Wirz, D. O., Heymann, H. & Bisson, L. F. (2004) Descriptive analysis of Brettanomyces-infected
Cabernet Sauvignon wines. American Journal of Enology and Viticulture. 55, 303A.
Wood, C., Siebert, T. E., Parker, M., Capone, D. L., Elsey, G. M., Pollnitz, A. P., Eggers, M.,
Meier, M., Vössing, T., Widder, S., Krammer, G., Sefton, M. A. & Herderich, M. J. (2008)
From wine to pepper: Rotundone, an obscure sesquiterpene, is a potent spicy aroma
compound. Journal of Agricultural and Food Chemistry. 56, 3738 - 3744.
62
Chapter 4: 5BSensory profiling of four separate Brett-related compounds in Pinotage red wine: 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol
and isovaleric acid
1 INTRODUCTION............................................................................................................. 63
2 MATERIALS AND METHODS........................................................................................ 64
2.1 Wine samples.............................................................................................................. 64
2.2 Chemicals and spiking............................................................................................... 64
2.3 Singular profiling of samples .................................................................................... 66
2.4 Data analysis............................................................................................................... 68
3 RESULTS AND DISCUSSION ....................................................................................... 68
3.1 4-ethylphenol .............................................................................................................. 68
3.2 4-ethylguaiacol ........................................................................................................... 72
3.3 4-ethylcatechol ........................................................................................................... 75
3.4 Isovaleric acid............................................................................................................. 78
3.5 Overall discussion of common descriptors............................................................. 81
4 CONCLUSIONS.............................................................................................................. 82
5 REFERENCES................................................................................................................ 83
63
1 INTRODUCTION
Brettanomyces yeast causes the wine defect commonly known as Brett character or phenolic
off-flavour, which is associated with an aroma that can be described as horsey, leathery,
medicinal, smoky or savoury (Wirz et al., 2004; Norris, 2004). Although the microbiological
characteristics of this yeast have been exhaustively studied, the current understanding of its
sensory effects in wine is rather limited. Two compounds, namely 4-ethylphenol and 4-
ethylguaiacol are generally accepted to be mainly responsible for the sensory effects associated
with Brettanomyces, and their presence in wine is used as diagnostic criterion for this type of
spoilage. 4-ethylphenol is associated with leather-like and Elastoplast™ descriptors, whereas 4-
ethylguaiacol has been linked to medicinal, spicy and clove like descriptors (Chatonnet et al.,
1992).
Isovaleric acid was pointed out by Licker et al. (1999) as one of the most odour-active
substances with regard to Brett character. However, some authors have found contrasting
results, as no significant difference in isovaleric acid level could be found between wines
inoculated with Brettanomyces and a control (Fugelsang & Zoecklein, 2003). More recently,
Romano et al. (2009) found a significant correlation between isovaleric acid levels and sensory
Brett character, and speculated that the rancid/pungent aroma of this compound may contribute
to what is described as Brett character.
The sensory character of 4-ethylcatechol has been described as “horsey” (Hesford &
Schneider, 2004) and smoky (Larcher et al., 2008). It has, however, been found that 4-
ethylcatechol does not have as an intense or recognisable sensory effect as the other volatile
phenols, and it has been speculated that its sensory effect is mainly due to synergism with the
other Brett-related compounds (Larcher et al., 2008).
Although several studies to date have investigated the sensory effects of 4-ethylphenol
and 4-ethylguaiacol (Eteiévant et al., 1989; Chatonnet et al., 1992; Cliff & King, 2009), no formal
sensory investigation has been undertaken that includes both these compounds as well as 4-
ethylcatechol and isovaleric acid. Obtaining a better understanding of their sensory effects when
present individually is the preliminary step to obtaining a better understanding of their sensory
interactions.
The aim of this study is therefore to investigate the separate sensory effects if 4-
ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid when present in Pinotage red
wine at varying levels. These four compounds were selected or further investigation as they
were the compounds most commonly linked to Brettanomyces spoilage in literature. This study
served as an exploratory step for the profiling of these compounds in combination, which will be
presented in Chapter 5.
64
2 MATERIALS AND METHODS
2.1 22BWine samples
Two hundred litres of Pinotage red wine was obtained from a local producer (Distell Group Ltd,
Stellenbosch, South Africa) during the course of 2009 and was bottled manually at the
Department of Viticulture and Oenology, Stellenbosch University, South Africa. The wine was
made using standard red wine making practices and completely underwent both alcoholic and
malolactic fermentation. The wine had an alcohol concentration of 11.6 % and a pH of 3.65.
After bottling, samples were taken to determine the levels of 4-ethylphenol, 4-
ethylguaiacol, 4-ethylcatechol and isovaleric acid in the wine. The concentrations of 4-
ethylphenol, 4-ethylguaiacol and isovaleric acid were determined using gas chromatography
mass spectrometry (GC-MS) and the concentration of 4-ethylcatechol was determined using
HPLC-MS/MS. All these analyses were conducted by an accredited wine analysis laboratory
(Quantum Laboratories, South Africa) and the methods are described in detail in Chapter 6.
During the chemical analyses of the wine used during this study, it was found that the
wine contained less than 10 µg/L of 4-ethylphenol, less than 10 µg/L of 4-ethylguaiacol, less
than 10 µg/L 4-ethylcatechol, and 355 µg/L of isovaleric acid. From this data, the wine is
considered to be free of ethylphenols. The level of isovaleric acid is close to the minimum of 300
µg/L found in red wines, as quoted by Francis & Newton (2005). The wine is therefore
considered free from microbiological spoilage.
2.2 23BChemicals and spiking
Solutions of 10 mg/mL, 1 mg/mL, 100 µg/mL and 10 µg/mL of 4-ethylphenol, 4-ethylguaiacol, 4-
ethylcatechol and isovaleric acid (Aldrich, South Africa) were prepared in 99.5% ethanol (Merck
Chemicals, South Africa). These solutions were used to produce wine samples spiked with the
desired concentrations of each compound. This study focussed on the separate sensory effects
of the four compounds, and therefore wine samples were spiked with only one of the four
compounds.
The concentrations of the four compounds used were identical to those used in the
central composite design (see Chapter 5). In Chapter 5, the design and the method for
concentration selection is described in more detail. The choice to correspond the levels used in
the singular sensory profiling (Chapter 4) with those of the central composite design (Chapter 5)
was made so that the two datasets could be compared.
The concentrations used were predetermined as follows: Firstly, detection thresholds
were determined as described in Chapter 3, and it was decided that level 2 in the central
65
composite design should correspond to the determined detection threshold. This was done to
allow for the investigation of sensory effects of the compounds at, below, and above their
detection thresholds. The highest level (level 5) corresponded to the highest level that is likely to
occur naturally in wine. This level was decided on with the guidance of literature and following
consultation with wine-industry experts (Thales, South Africa). Levels 2 and 5 were used to
calculate the rest of the levels of the central composite design. Due to the inherent differences
between the base wines used in Chapter 3 and this chapter, all levels used were subjected to
extensive sensory pre-screening in order to determine whether they adhere to the sensory
criteria set.
The central composite design levels were chosen along the same logarithmic scale as
the eight levels used for the detection thresholds. The levels used during detection thresholds
all fall on the curve c = ab n-1 , where c, is the concentration, a is the starting point (level 1), b is
the multiplication factor and n is the level number. The levels chosen in the design were on a
simple numerical scale that corresponded to that used in the detection thresholds in they also
satisfy c = ab n-1. This is shown in XFigure 4.1 X. This was done to compensate for the fact that the
distances between levels in the central composite design are predetermined. This created a
challenge as the highest levels that this research wished to investigate were between five and
twenty times the magnitude of the detection threshold.
Figure 4.1. Graphical explanation of the method of level selection. Both these curves have the
equation c=ab n-1, where c is the concentration, b is the multiplication factor used, and n is the
level. a) Shows the levels used during the determination of detection thresholds (Chapter 3). b)
Shows the levels used in this chapter, falling along the same curve as those in a).
The use of a logarithmic scale compensated for the differences in magnitude, and
allowed for the concentrations in question to be fitted to the central composite design. For this
reason, two sets of levels are used in this study. The “design” levels are those determined by
the design, and fall between 0 and 13 ( XTable 4.1X). The actual concentrations used were
determined by logarithmically transforming these levels and are shown in XTable 4.2 X. In both
a)
1 2 3 45
6
7
8
0
2
4
6
8
10
12
0 2 4 6 8 10
Level
Co
nce
ntr
atio
n
b)
2.84.2
5.6
7
8.4
0
2
4
6
8
10
12
0 2 4 6 8 10
Level
Co
nce
ntr
atio
n
a)
1 2 3 45
6
7
8
0
2
4
6
8
10
12
0 2 4 6 8 10
Level
Co
nce
ntr
atio
n
b)
2.84.2
5.6
7
8.4
0
2
4
6
8
10
12
0 2 4 6 8 10
Level
Co
nce
ntr
atio
n
66
these tables, levels are numbered 1 to 5, and these values are used throughout this chapter
when referring to samples, with reference to the concentrations and design levels where
necessary.
Concentration ranges were finalised with the help of wine-industry experts by means of
several sessions of consensus sensory analysis. Before commencement of the formal sensory
tests, the final samples were also subjected to sensory pre-assessment, using the mentioned
wine-industry experts.
Table 4.1. Design levels for spiking of wines with Brett-related compounds.
Level Compound
1 2 3 4 5
4-ethylphenol 0.5 3.5 6.5 9.5 12.5
4-ethylguaiacol 0.75 2.5 4.5 6 7.75
4-ethylcatechol 2.8 4.2 5.6 7 8.4
Isovaleric acid 2.5 4.5 6.5 8.5 10.5
Table 4.2. Actual concentrations of Brett-related compounds tested.
Level Compound
1 2 3 4 5
4-ethylphenol (µg/L) 82 227 623 1711 4695
4-ethylguaiacol (µg/L) 65 117 230 381 688
4-ethylcatechol (µg/L) 181 290 465 745 1193
Isovaleric acid(µg/L) 381 431 577 997 2210
2.3 24BSingular profiling of samples
The compounds were profiled using quantitative descriptive analysis according to the general
descriptive method (Lawless & Heymann, 1998). The concentrations of the compounds are
listed in XTable 4.2X. The samples that were analysed contained the base wine and only one of
the compounds, therefore the samples were analysed in sets of six containing 5 spiked samples
plus one control sample (not being spiked). The panel consisted of 10 judges who had previous
experience in the use of quantitative descriptive analysis. Most of the panellists also took part in
the determination of detection thresholds for these compounds (Chapter 3) and therefore
already had a degree of familiarity with the compounds under investigation.
The first phase of training consisted of five sessions of general training. During the first
session, the judges received all the samples, as well as a range of reference standards adapted
from Noble et al. (1987). The purpose of this session was to familiarise the judges with the
samples and descriptors used during this part of the study. During the subsequent four training
67
sessions, the samples were evaluated per compound, and descriptors were generated for use
during quantitative descriptive analysis.
The second phase of training consisted of two session of training per compound
(therefore eight sessions in total). During the first training session, descriptors were finalised,
and an attempt was made to finalise the positions of the samples on a 100 mm-unstructured
line scale. During the second training session, the positions of the different samples on the
unstructured line scale were finalised, and an attempt was made to reach consensus regarding
the different samples.
The descriptors used in the final descriptive analysis, as well as the reference standards
used to define them, are shown in XTable 4.3X.
Table 4.3. Reference standards and descriptors used during singular profiling of wines spiked
with variable concentrations 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid.
Descriptor Definition Reference standard Compound relevant for descriptor
Berry-like Typical wine-sweet berry-like aroma
Control samples All
Sick-sweet Atypical wine-sweet – sweet smell that is uncharacteristic/unnatural for wine
None All
Elastoplast™a Smell associated with Band-Aid™or Elastoplast™
1 piece of Elastoplast™ bandage 4-ethylphenol
Leather The smell associated with leather
1 piece of leather 4-ethylphenol
Smoky Smoky smell associated with wine
Wine spiked with “smoke essence” used in meat products (3 drops in 100 mL wine)
4-ethylguaiacol
Medicinal/
Listerine-like
Minty smell associated with mouthwash
Wine spiked with Listerine™, a mouthwash (3 drops in 100 mL wine)
4-ethylguaiacol
Savoury Meaty smell associated with food
Wine spike with soy sauce (1 mL in 100 mL wine)
4-ethylcatechol
Pungent Sweaty/ rancid/ cheesy/ vinegary
A small piece of mild blue cheese (Simonzola™)
Isovaleric acid
aThroughout this study, the term “Elastoplast™” is used. This term is the local equivalent of “Band-Aid™”.
Each compound was analysed during four sessions of descriptive analysis (16 session
in total). The final profiling analyses were conducted by 10 trained assessors in booths with
standard artificial daylight lighting and temperature control at 20°C ±1°C. The wine was
analysed in standard ISO wine tasting glasses, sample size was 20 mL and samples were
served at 20°C ±1°C. Prior to the analysis, the wine glasses containing the wine samples were
68
covered with plastic lids. This prevented the aroma of the wine from escaping or contaminating
the laboratory environment. Samples were marked with a random three-digit code, and were
randomised within the judges. However, the non-spiked control sample was labelled “C”, and
was always presented in the first position.
2.4 25BData analysis
A randomized complete block design was used for the sensory analysis where each judge
received a control sample containing only the base wine and five spiked samples. This was
replicated four times. The data were analysed using SAS® software (Version 9; SAS Institute
Inc, Cary, USA) and subjected to the Shapiro-Wilk test for non-normality of the residuals
(Shapiro & Wilk, 1965). If non-normality was found to be significant (P≤0.05) and caused by
skewness, the outliers were identified and removed until the data were normal or symmetrically
distributed (Glass et al., 1972). Using line plots indicating temporal stability and internal
consistency, single odd judges were identified and removed. The final analysis of variance
(ANOVA) was performed after the above-mentioned procedures. Student’s t-least significant
difference (LSD) was calculated at the 5% significance level to compare treatment means.
Discriminant analysis and Principal Component Analysis (PCA) were performed on responses
for the different judges of the different levels. Multivariate data analyses were performed using
the XLStat software (Version 2009.5.0.1, Addinsoft, SARL, Paris, France).
3 RESULTS AND DISCUSSION
3.1 26B4-ethylphenol
The overall trends found during the profiling of 4-ethylphenol are shown in XFigure 4.2 X, and the
mean values with least significant differences are shown in XTable 4.4 X. As can be seen in this
figure and table, there was a large decrease in berry-like character when 4-ethylphenol is added
at levels below its detection threshold (level 2). There was also a significant increase in the
levels of the Elastoplast™ and sick-sweet descriptors. However, there were no significant
changes between the first addition, detection threshold and the level above detection threshold,
in terms of the sick-sweet and Elastoplast™ descriptors. The berry-like character continued to
decrease up to level 4 (1711 µg/L). The sick-sweet and Elastoplast™ descriptors increased
significantly from level 3 (623 µg/L) to level 4 (1711 µg/L), and from level 4 (1711 µg/L) to level
5 (4695 µg/L). In terms of the leather descriptor, there was no real difference between the non-
spiked control samples and additions of level 1 (82 µg/L), 2 (227 µg/L), or 3 (62 3µg/L).
69
Although there was a significant difference (p ≤ 0.05) between the non-spiked sample and level
2, there was no difference between this sample and level 1, and level 3 and level 1. However,
the level of leather character increases significantly from an addition of level 3 to level 4, and
from level 4 to level 5.
0
10
20
30
40
50
60
70
0 2 4 6 8 10 12 14Level of 4-ethylphenol
Inte
ns
ity
of
de
sc
rip
tor
Elastoplast
Leather
Berry-like
Sick-sweet
Figure 4.2. Change in sensory profile of Pinotage wine due to the addition of 4-ethylphenol.
Levels are “design levels”.
Table 4.4. Change in sensory profile of Pinotage wine due to the addition of 4-ethylphenol.
Level (conc) Berry-like1 Sick-sweet1 Elastoplast™1 Leather1
0(0 µg/L) 58.8 a 0.6 e 0.4 d 0.3 d
1(82 µg/L) 32.0 b 9.8 d 8.6 c 2.8 c d
2(227 µg/L) 25.4 c 14.8 c 11.8 c 5.4 c
3(623 µg/L) 26.0 c 12.4 c d 11.6 c 3.2 d
4(1711 µg/L) 8.3 d 29.9 b 33.8 b 15.7 b
5(4695 µg/L) 5.9 d 34.6 a 46.6 a 23.4 a
Least Significant Difference* (p = 0.05)
4.56 3.65 4.70 3.20
1 Values with the same superscript are not significantly different (p = 0.05).
From this it can be assumed that the effect of 4-ethylphenol manifests itself as a
suppression of natural berry-like character below level 4, which results in a sick-sweet
character. The typical Elastoplast™ and leather-like aromas associated with this compound are
only significantly present from concentrations higher than level 4 (1711 µg/L).
XFigure 4.3X shows the results of the discriminant analysis performed on the different
levels of 4-ethylphenol. The first two factors explained 99% of the variance. XFigure 4.3 X shows a
clear separation between levels 4 and 5 and level 0, the non-spiked sample. There is also a
70
separation between level 0 and level 3. From this it can be seen that judges could detect a
difference with an addition of level 3 of 4-ethylphenol, but that this difference only became clear
from level 4. This substantiates the results found during the univariate analysis of the data.
0000
0
000
0
0
1
1
1
1
11 1
1
11
2 2
2
22
2
2
2
2
2
33
3
L3
3
3
3
3 3
3
444
4
44
4
4
44
5
5
55
5
5
5
5
5
5
5
4
3
21
0
-6
-4
-2
0
2
4
6
-6 -4 -2 0 2 4 6
F1 (78.64 %)
F2
(19.
88 %
)
0
1
2
3
4
5
Centroids
Level
Figure 4.3. Discriminant analysis of data obtained during sensory analysis of wines spiked with
4-ethylphenol.
XFigure 4.4X shows a PCA biplot of the results obtained during profiling of the samples
spiked with 4-ethylphenol. 96% of the variation was described by the first two factors, with 87
and 9% of the variation being described by factors F1, and F2 respectively. This means that the
samples have a more or less linear distribution along F2. It can be seen that F1 is driven by the
different descriptors, with berry-like negatively associate with F1, and Elastoplast™, sick-sweet
and leather positively associated with F1. Along F2, the loadings for Elastoplast™ and leather
are grouped together, but are separated from sick-sweet. This can be ascribed to the slightly
different pattern that the sick-sweet descriptor followed in XFigure 4.2 X. This aroma increases
more drastically with the addition of lower levels of 4-ethylphenol than the other descriptors, but
does not show as large an increase above level 4. The loading for sick-sweet also shows a
strong negative association to the berry-like descriptor along F1, which indicates that it arises as
a result of the suppression of the natural berry-like character of the wine.
71
5
5555
5
5
5
55
4
44
4444
4
4
4
3 333
33
3
33 3
22
2
2 2
2
2
2
2
2
1
11
11 11
1
11
00000000 00 ElastoplastLeather
Sick-sweet
Berry-like
-5
-4
-3
-2
-1
0
1
2
3
4
5
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8F1 (86.81 %)
F2
(8.9
0 %
)
Figure 4.4. PCA biplot of all data obtained during sensory of samples containing different levels
of 4-ethylphenol. Level of 4-ethylphenol is indicated by corresponding numbers.
The scores for the different levels of 4-ethylphenol also separate along F1 – levels 0 to 3
of 4-ethylphenol fall to the negative (left) side of F1, and levels 4 and 5 tend to fall to the positive
(right) side of F1. From this it can be concluded that the leather, Elastoplast™ and sick-sweet
characteristics are more prominent in levels 4 and 5, as these scores associate with the
loadings associated with these attributes. This yet again substantiates the results found during
the univariate statistical analyses, as the levels of 4-ethylphenol (levels 4 and 5) which
exhibited high levels of the leather, Elastoplast™ and sick-sweet descriptors associate with the
loadings for those descriptors (XTable 4.4X). The levels which did not show significant differences
from one another in all the attributes (levels 1, 2 and 3) also associate with one another in
XFigure 4.4X. Significant differences were however found between these samples and levels 4 and
5. This explains the separation of these samples along F1 in XFigure 4.4X.
The overall increase in the Elastoplast™ and leather descriptors with an increase in 4-
ethylphenol concentration is expected, as these descriptors have been linked to this compound
by several other authors (Chatonnet et al., 1992; Curtin et al., 2008). However, the “stable” and
“barnyard” descriptors that have also been linked to this descriptor could not be perceived or
quantified by the panel. This may be due to two reasons. Firstly, it is likely that these aromas
could not be recognised by the panel due to a lack of internal reference for these aromas
caused by a lack of exposure to these odours (Hughson & Boakes, 2002). It is also possible
that the “stable” and “barnyard” descriptors are a result of interaction of different compounds
72
associated with this defect. Although such blending phenomena have not yet been explored in
wine, they have been found by Brodin et al. (2007) and Le Berre et al. (2008) in other odourant
mixtures.
3.2 27B4-ethylguaiacol
The overall results obtained during the profiling of Pinotage spiked with 4-ethylguaiacol are
shown in XFigure 4.5 X. The mean values of the samples and their least significant differences are
shown in XTable 4.5X.
Table 4.5. Change in sensory profile of Pinotage wine due to the addition of 4-ethylguaiacol.
Level(conc) Berry-like1 Sick-sweet1 Medicinal1 Smoky1
0 (0 µg/L) 59. 4 a 1. 9 e 1. 6 d 3. 3 e
1 (65 µg/L) 35. 2 b 12. 9 b c 5. 5 c 12. 6 c d
2 (117 µg/L) 33. 9 b 9. 1 d 6. 2 c 10. 1 d
3 (230 µg/L) 31. 8 b 10. 4 c d 6. 4 c 17.0 c
4 (381 µg/L) 17.5 c 15. 7 b 11. 8 b 28. 7 b
5 (688 µg/L) 9. 9 d 20. 8 a 17. 2 a 44. 3 a
Least Significant Difference (p=0.05)
5.24 3.36 2.30 5.75
1 Values with the same superscript are not significantly different (p = 0.05).
As is evident in XFigure 4.5 X and XTable 4.5 X, a significant decrease in berry-like character
occurred with an addition of 4-ethylguaiacol below its detection threshold. However, the next
significant change in berry-like character only occurs at level 4 (381 µg/L), which implies that the
change in this character only occurred in Pinotage red wine when the compound’s
concentration was significantly above it detection threshold. Similarly, a significant increase in
the medicinal and smoky attributes can be observed at an addition below detection threshold.
Although but there is no significant increase from below to above detection threshold, a further
significant change is observed from level 3 (230 µg/L) to level 4 (381 µg/L). The sick-sweet
descriptor follows a similar pattern, although here the difference occurs from addition below
detection threshold to level 5 (688 µg/L). This means that the presence of 4-ethylguaiacol
induced the sick-sweet attribute of the wine, but this attribute only increased in intensity when
this compound was present at very high levels.
Discriminant analysis performed on the different levels of 4-ethylguaiacol shows a clear
separation between level 0 and levels 3, 4 and 5 ( XFigure 4.6X). This makes sense as there were
no real significant differences in the different descriptors between the other levels, which
indicates that the judges could not clearly perceive differences between the samples. For this
reason, the samples are not expected to separate clearly on XFigure 4.6 X, as the differences are
too small to cause clear separation.
73
0
10
20
30
40
50
60
70
0 1 2 3 4 5 6 7 8 9
Level of 4-ethylguaiacol
Inte
nsi
ty o
f d
escr
ipto
r
Medicinal
Smoky
Berry-like
Sick-sweet
Figure 4.5. Change in sensory profile of Pinotage wine due to an induced increase in 4-
ethylguaiacol level. Levels are “design levels”.
000000
0
00
1
1
1
1
1
1
1
1
1
22
22
22
2
2
2
33
33
3
3
3
33
44
4
4
4
4
4
4
4
5
5 5
5
5
5
5
5
5
5
4 3
2
1
0
-4
-3
-2
-1
0
1
2
3
4
-4 -3 -2 -1 0 1 2 3 4
F1 (88.41 %)
F2
(9.7
5 %
)
0
1
2
3
4
5
Centroids
Level
Figure 4.6. Discriminant analysis of data obtained during singular profiling of wines spiked with
4-ethylguaiacol.
XFigure 4.7X shows a PCA biplot for 4-ethylguaiacol. 87% of the total variation could be
explained by this biplot, 74% by F1 alone. This implies that the difference between samples is of
74
a near-linear nature. Similarly to what was found with 4-ethylphenol, F1 separates berry-like
from the other descriptors. This is due to the fact the berry-like character decreased with an
increase in concentration of 4-ethylguaiacol, whereas the other descriptors increased with an
increase in concentration. F2 separates the medicinal attribute from the sick-sweet and smoky
attributes. The smoky attribute associates with the sick-sweet attribute. Although this seems
unexpected, both these compounds follow similar patterns in XFigure 4.5 X, which explains their
association. As in the case of 4-ethylphenol, there is a strong negative association between the
loadings for sick-sweet and berry-like descriptors, which further substantiates the hypothesis
that the sick-sweet characteristic arises due to a suppression of the natural berry-like character
of the wine.
00
000000 0
11
1
1
1
11
1 1
2
2
2
2
22 2
2
2
3
3
3
33
3 3
3
3
4
4
4
44
4
4
4
4
5
5
5
5
5
5555
Berry-like
Smoky
Medicinal
Sick-sweet
-5
-4
-3
-2
-1
0
1
2
3
4
5
-5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
F1 (73.65 %)
F2
(13.
19 %
)
Figure 4.7. PCA biplot of all data obtained during sensory of samples containing different levels
of 4-ethylguaiacol. Level of 4-ethylguaiacol is indicated by corresponding numbers PCA biplot of
samples for 4-ethylguaiacol
In terms of sample scores, levels 0, 1, and 2 tend to fall to the negative side of F1, and
samples 4 and 5 tend to fall towards the positive side. However, level 3 tends to fall towards the
middle of the biplot, which fits in with levels 0, 1, and 2 being strongly associated with the berry-
like attribute, but not with the medicinal or smoky attributes. Conversely, levels 4 and 5 were
associated with the medicinal and smoky attributes, which explains their positions. The
distribution of level 3 can be explained by looking at XFigure 4.6 X, which indicates a poor
discrimination between this level and all the other levels except level 0.
75
Overall, the increase in the medicinal aroma caused by an increase in 4-ethylguaiacol is
expected, as this aroma has been linked with Brett character in literature (Curtin et al., 2008).
However, the clove-like or spicy aroma that is commonly linked to this compound (Chattonet et
al., 1992; Licker et al., 1999; Curtin et al., 2008) could not be accurately identified in the panel.
This was in spite of the fact that the reference standard for the clove-like aroma (Noble et al.,
1987) was employed. This may be due to the fact that the Medicinal and the clove-like
descriptors were cognitively similar, which prevented the panel from distinguishing between
discriminating between these aromas (Escudero et al., 2007). Finally, the smoky attribute
coupled to this compound in this study has not been previously linked with 4-ethylguaiacol or
wines spoiled with Brettanomyces. This is further indication of the sensory complexity of Brett
character.
3.3 28B4-ethylcatechol
A summary of the results obtained for 4-ethylcatechol are shown in XTable 4.6 X and XFigure 4.8 X.
0
10
20
30
40
50
60
70
0 1 2 3 4 5 6 7 8 9
Level of 4-ethylcatechol
Inte
ns
ity
of
de
sc
rip
tor
Savory
Berry-like
Sick-sweet
Figure 4.8. Effect of a change in 4-ethylcatechol level on the sensory profile of Pinotage wine.
Levels are “design levels”.
XFigure 4.8X and XTable 4.6 X show that the addition of 4-ethylcatechol decreases the berry-
like character of a wine when it is added below detection threshold, but does not significantly
affect this attribute with further additions. The sick-sweet characteristic follows a similar pattern,
but there is a significant difference between an addition below detection threshold (290 µg/L)
and levels 4 (745 µg/L) and 5 (1193 µg/L). In terms of savoury character, the addition below
76
detection threshold also shows a significant change, but levels 4 and 5 are significantly different
from additions below detection threshold. The results seem to indicate that 4-ethylcatechol on
its own not does significantly affect the overall profile of the wine to the same degree as 4-
ethylphenol and 4-ethylguaiacol, as predicted by Larcher et al. (2008). Considering that the a
100 mm scale was used, the change in berry-like character is approximately 35 out of a
possible 100 – which is a significant change but still significantly less than the 50 change
caused by the addition of 4-ethylphenol and 4-ethylguaiacol. In addition, the maximum levels of
the sick-sweet and savoury characteristics were approximately 10 out of a possible 100, which
is generally termed as just detectable.
XFigure 4.9X shows the results for the discriminant analysis of the data for 4-ethylcatechol.
Level 0 separates from levels 4 and 5, but the rest of the levels neither separate from one
another nor from Level 0. This implies that the panel experienced difficulty in judging the subtle
differences between the lower levels of 4-ethylcatechol. This configuration is also mirrored by
the groupings in terms of least significant differences shown in XTable 4.6
A PCA biplot of the data obtained during the profiling of 4-ethylcatechol is shown in
XFigure 4.10 X. 94% of the variation is explained by the first two components, F1 and F2 account
for 83 and 11% respectively. Similar to 4-ethylphenol and 4-ethylguaiacol, F1 separates the
berry-like descriptor from the savoury and sick-sweet attributes. In XFigure 4.10 X, the sick-sweet
and savoury attributes are differentiated along F2. A strong negative association between the
sick-sweet and berry-like attributes (like in the cases of 4-ethylphenol and 4-ethylguaiacol) could
not be observed. However, when looking at XFigure 4.8 X, it can be seen that the decrease in
berry-like character is much more severe than the increase in sick-sweet character, where in the
cases of 4-ethylphenol and 4-ethylguaiacol there was a similar substantial increase. This may
indicate that although 4-ethylcatechol suppresses berry-like character in wine, its elusive
character involves interaction with the production of sick-sweet aromas through the suppression
of berry-like character. Although the scores are arranged in a linear-like fashion, the only clear
pattern that can be seen is the separation of the non-spiked samples from the spiked samples.
This substantiates the pattern found in the discriminant analysis and the univariate analyses.
Table 4.6. Change in sensory profile of Pinotage wine due to changes in 4-ethylcatechol level.
Level(conc) Berry-like1 Sick-sweet1 Savoury1
0 (0 µg/L) 58.9 a 0.3 d 0.5 c
1 (181 µg/L) 30.4 b 7.0 c 7.6 b
2 (290 µg/L) 29.9 b 7.5 b c 9.6 a b
3 (465 µg/L) 29.7 b 8.1 a b c 7.5 b
4 (745 µg/L) 25.6 b 11.0 a 10.6 a
5 (1193 µg/L) 25.3 b 10.1 a b 11.4 a
Least Significant Difference (p=0.05)
5.35 2.97 2.69
1 Values with the same superscript are not significantly different (p = 0.05).
77
0L00
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
3
3
3
33
3
3
3
3
4
4
4
4
4
4
4
44
5
55
5
5
5
5
5
5
0
1
2
3
45
-4
-3
-2
-1
0
1
2
3
4
-4 -3 -2 -1 0 1 2 3 4
F1 (91.20 %)
F2
(6.4
7 %
)
0
1
2
3
4
5
Centroids
Level
Figure 4.9. Discriminant analysis of all data obtained during sensory profiling of Pinotage spiked
with different levels of 4-ethylcatechol.
5
555
5
4
4
4
4
44
44
4
3 3
333
3
3
3
3
22
22
22
2 2
2 11
11
1
1
1
1
1
00 00 00 000
Savory
Berry-like
Sick-sweet
-3
-2
-1
0
1
2
3
-5 -4 -3 -2 -1 0 1 2 3 4 5
F1 (83.10 %)
F2
(11.
30 %
)
Figure 4.10. PCA biplot of all data obtained during sensory profiling of samples containing
different levels of 4-ethylcatechol. Level of 4-ethylcatechol is indicated by corresponding
numbers.
78
It is interesting to note that neither of the descriptors linked to 4-ethylcatechol in literature
– namely “smoky” (Larcher et al., 2008; Curtin et al., 2008) and “horsey” (Hesford et al., 2004) –
were found during this study. However, the savoury attribute could be similar to the “soy”
descriptor that was coupled to Brettanomyces-infected wines by Wirz et al. (2004). The savoury
attribute is closely related to all three these attributes. The findings of this study regarding the
sensory effects of 4-ethylcatechol can therefore be regarded to be in line with literature.
3.4 29BIsovaleric acid
In a similar fashion to the other compounds, a significant decrease in berry-like character was
observed when isovaleric acid was added to the wine at low concentrations (XFigure 4.11 X and
XTable 4.7X). However, the only other significant decrease in berry-like character was seen at an
addition of level 5 (2210 µg/L). The sick-sweet character also increased from an addition below
detection threshold, with the next two significant changes occurring at levels 4 (997 µg/L) and 5
(2210 µg/L). The pungent aroma also increased significantly with an addition below the
detection threshold, and again increased significantly with an addition of levels 4 and level 5.
XFigure 4.12X shows the discriminant analysis results for the sensory profiling for isovaleric
acid. Level 0 is separated from all the levels except for level 2. Although none of the other levels
are separated on this figure, near separation is observed between level 3 and 5. These two
classes are spanned by level 4. This implies that although levels 3 and 5 are different, level 4
shares properties with both these levels. This may have been caused by the lack of significant
difference between level 3 and level 4 in terms of both the berry-like and sick-sweet
characteristics.
Table 4.7. Change in sensory profile due to an increase in isovaleric acid.
Level(conc) Berry-like1 Sick-sweet1 Pungent1
0 (355 µg/L) 60.0 a 0.1 d 0.0 d
1 (381 µg/L) 35.7 b 9.4 c 13.0 c
2 (431 µg/L) 34.9 b 6.9 c 11.9 c
3 (577 µg/L) 37.5 b 9.9 b c 13.0 c
4 (997 µg/L) 33.9 b 12.7 b 18.7 b
5 (2210 µg/L) 26.2 c 15.8 a 25.6 a
Least Significant Difference (p=0.05)
4.45 3.03 3.92
1 Values with the same superscript are not significantly different (p = 0.05).
79
0
10
20
30
40
50
60
70
0 2 4 6 8 10 12
Level of isovaleric acid
Inte
ns
ity
of
des
cri
pto
r
Pungent
Berry-like
Sick-sweet
Figure 4.11. Change in sensory profile due to the increase in isovaleric acid. Levels are “design
levels”.
The results of PCA performed on the data obtained during the profiling of isovaleric acid
can be seen in XFigure 4.13 X. 94% of the variance could be explained by the first two
components, with F1 explaining 85% of the variation and F2 explaining 9%. Similar to the other
compounds, F1 is characterised by the loading for berry-like on the negative side and sick-
sweet and pungent on the positive side. F2 is characterised by the separation of the sick-sweet
and berry-like attributes from the pungent attribute. It is interesting to note that berry-like and
sick-sweet do not have the strong negative association that was shown in the case of 4-
ethylphenol and 4-ethylguaiacol. However, the pungent and berry-like attributes do show a
strong negative association. This makes some sense as the lines for these two attributes in
XFigure 4.11 X seem to be mirror-images of one another – a decrease in the berry-like attribute is
associated with a comparable increase in the pungent attribute. This inversion of aroma
character compared to the other compounds may be attributed to the overpowering nature of
the pungent attribute, making it difficult for the judges to identify the sick-sweet characteristic in
these samples.
The scores for level 1, 2 and 3 are distributed all over the biplot, and do not associate
with either of the other levels or each other. This is probably because of the lack of clear
difference between these three samples, which could be seen from the univariate results.
80
L0L00L0L0L0000
1
1
1
1
1
1
1
11
2
2
2
2
2
22
2
2
3
3
33
3
3
3
3
3
4
444
4
4
4
4
55
55
5
55
0
1
2
34 5
-6
-4
-2
0
2
4
6
-6 -4 -2 0 2 4 6
F1 (92.08 %)
F2
(6.9
6 %
)
0
1
2
3
4
5
Centroids
Level
Figure 4.12. Results of discriminant analysis performed on all data obtained during the sensory
profiling of Pinotage spiked with different levels of isovaleric acid.
000000000
11
11
1
1
1
1122
2 2
2
2 222 33
33
3
33
3
3
4
444
4
4
4 4
5
5
55
5
5
5
Berry-like Sick-sweet
Pungent
-3
-2
-1
0
1
2
3
-5 -4 -3 -2 -1 0 1 2 3 4 5
F1 (85.20 %)
F2
(8.5
6 %
)
Figure 4.13. PCA biplot of all data obtained during sensory of samples containing different
levels of isovaleric acid. Level of isovaleric acid is indicated by corresponding numbers.
81
Previous investigations into the sensory effects of isovaleric acid in wine linked this
compound with “rancid” (Licker et al., 1999; Romano et al., 2008) and “sweaty” (Romano et al.,
2009) descriptors. The pungent attribute used in this study aimed to include these two
descriptors. The sensory results of this study regarding isovaleric acid are therefore in line with
that which is expected from literature.
3.5 30BOverall discussion of common descriptors
Several general trends could be observed in terms of the sensory effect of these four Brett-
related compounds. Firstly, all four compounds caused suppression in berry-like character when
they were added at concentrations below their detection thresholds. Secondly, the only
significant difference in berry-like character between level 1 (below detection threshold) and
level 2 (detection threshold) were found for 4-ethylphenol. However, there were no significant
changes in berry-like character between level 2 (approximately detection threshold) and level 3
(above detection threshold) for any of the compounds. This might be due to the fact that a
different Pinotage wine was used in this study than in Chapter 3.
In some of the cases (4-ethylphenol and 4-ethylguaiacol), there was a significant
decrease in berry-like character from level 3 to level 4, and in some cases (4-ethylguaiacol and
isovaleric acid) a significant decrease from level 4 to level 5. In the case of 4-ethylcatechol,
there was no significant change between level 2 and level 5. From these results it can be
concluded that the suppressant effect on berry-like character occurs most strongly at the lower
levels of 4-ethylphenol, but at the higher levels of 4-ethylguaiacol and isovaleric acid. Although
its presence caused a suppression of berry-like character, 4-ethylcatechol does not seem to
have a severe effect on the suppression of berry-like character with an increase in
concentration.
The sick-sweet descriptor followed an opposite general trend to the berry-like descriptor
for all compounds: There was an increase in sick-sweet character at additions below detection
threshold, but not always a significant increase to detection threshold (level 2) or from level 2 to
level 3 (detection threshold to above detection threshold). 4-ethylguaiacol did not follow this
pattern strictly, as the mean for this attribute at level 2 was significantly higher than that of level
3. This can be ascribed to the fact that the medicinal (or sometimes minty) character of 4-
ethylguaiacol could easily be mistaken for sick-sweet, causing interference in the analysis of this
descriptor. For the other three compounds, there were significant differences between level 1
and level 4 in all the cases, with further significant differences between levels 4 and 5 for 4-
ethylphenol, 4-ethylguaiacol and isovaleric acid. 4-ethylphenol caused the greatest increase in
sick-sweet character, whereas 4-ethylcatechol only caused slight increases.
82
In terms of the other descriptors – where different descriptors are relevant for different
compounds – it is interesting to note that most of the descriptors followed a similar trend to the
sick-sweet descriptor (and the opposite of the berry-like descriptor): an increase from level 0 to
level 1, no increase between levels 1 and 3, and an increase from level 3 to 4 and/or level 4 to
5. From these data it can be concluded that the berry-like descriptor character decreases with
increase of the “taint” descriptors.
4 CONCLUSIONS
As expected from literature (Licker et al., 1999; Fugelsang & Zoecklien, 2003; Ugarte et al.,
2005; Fariña et al., 2007; Cliff & King, 2009), the four different Brett-related spoilage
compounds all suppressed berry-like character in wines, although not to an equal extent. The
increase in the relevant descriptors for 4-ethylphenol, 4-ethylguaiacol and isovaleric acid were
expected, as these compounds had been linked with these attributes in literature (Chatonnet et
al., 1992; Licker et al., 1999; Curtin et al., 2008; Romano et al., 2009). The slight increase in
savoury character caused by 4-ethylcatechol is in agreement with the findings of Larcher et al.
(2008) and Hesford et al. (2004).
When profiling these components by themselves, it was found that all four compounds
suppressed berry-like character, and caused an increase in an atypical sick-sweet (or
confected) character. In the cases of 4-ethylphenol and 4-ethylguaiacol, it can be concluded
from the results of multivariate statistics (XFigure 4.4 X and XFigure 4.7 X) that this attribute is a direct
result of the suppression of fruitiness. Although this is also evident for 4-ethylcatechol and
isovaleric acid, this conclusion cannot be drawn for these compounds, as other factors seem to
influence the presence of the sick-sweet characteristic. As this attribute has not been coupled to
Brettanomyces spoilage to date, it could be speculated that although it is present with the
individual compounds, sensory interactions between these compounds negate this effect.
Several other authors (Chatonnet et al., 1993; Fugelsang & Zoecklein, 2003; Curtin et al., 2008;
Romano et al., 2009) have also speculated about the sensory interactions evident between
these compounds. This requires further investigation.
In this study, PCA has shown itself to be a more powerful tool for the interpretation of
sensory data than normal univariate plots and least significant differences, as it provided insight
into the relationships between different samples and attributes. Although the nature of the
experiments in this study were relatively simple, with one principal component explaining the
majority of variance in all the cases, the use of PCA was still insightful for the overall
interpretation. The exploratory use of PCA in sensory data is therefore recommended as a
complementary tool for univariate techniques in future studies.
83
5 REFERENCES
Brodin, M., Moeller, P. & Olsson, M. (2007) Are we mixing odorants or odors? Chemical
Senses. 32, A19.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
Cliff, M. A. & King, M. C. (2009) Influence of serving temperature and wine type on perception of
ethyl acetate and 4-ethylphenol in Wine. Journal of Wine Research. 20, 45 - 52.
Curtin, C. D., Bellon, J. R., Coulter, A. D., Cowey, G. D., Robinson, E. M. C., de Barros Lopes,
M. A., Godden, P. W., Henschke, P. A. & Pretorius, I. S. (2005) The six tribes of "Brett"
in Australia - Distribution of genetically divergent Dekkera bruxellensis strains across
Australian winemaking regions. Australian and new Zealand Wine industry Journal. 20,
28 - 35.
Eteiévant, P. X., Issanchou, S. N., Marie, S., Ducruet, V. & Flanzy, C. (1989) Sensory impact of
volatile phenols on red wine aroma: influence of carbonic maceration time and storage.
Sciences des aliments. 9, 19 - 31.
Fariña, L., Boido, E., Carrau, F. & Dellacassa, E. (2007) Determination of volatile phenols in red
wines by dispersive liquid-liquid microextraction and gas chromatography-mass
spectrometry detection. Journal of Chromatography A. 1157, 46 - 50.
Francis, I. L. & Newton, J. L (2005) Determining wine aroma from compositional data. In: AWRI
– Advances in wine science (edited by R. J. Blair, M. E. Francis & I. S. Pretorius A. L.),
Pp 201 - 212. Glen Osmond, Australia: Australian Wine Research Institute.
Fugelsang, K. C. & Zoeklein, B. W. (2003) Population dynamics and effects of Brettanomyces
bruxellensis strains on Pinot noir (Vitis vinifera L.) wines. American Journal of Enology
and Viticulture. 54, 294 - 300.
Glass, G.V., Peckham, P.D. & Sanders, J.R. (1972). Consequences of failure to meet
assumptions underlying the fixed effects analyses of variance and covariance. Review of
Educational Research, 42, 237 - 288.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11 - 13.
Hughson, A. L. & Boakes, R. A. (2002) The knowing nose: the role of knowledge in wine
expertise. Food Quality and Preference. 13, 463 - 472.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Lawless, H. T. & Heymann, H. (1998). Sensory evaluation food, principles and practice. New
York, USA: Chapman and Hall.
84
Licker, J. L, Acree, T. E. & Henick-Kling, T. (1999) What is "Brett" (Brettanomyces) flavour? A
preliminary investigation. In: Chemistry of Wine Flavour. ACS Symposium Series (edited
by A. L. Waterhouse & S. E. Ebeler), Pp 96 - 115. Washington DC: American Chemical
Society.
Noble, A.C., Arnold, R.A., Buechsenstein, J., Leach, E. J., Schmidt, J. O. & Stern, P. M. (1987)
Modification of a standardized system of wine aroma terminology. American Journal of
Enology and Viticulture. 38, 143 - 146.
Norris, L. (2004) Unraveling the mystery of Brettanomyces flavor. American Journal of Enology
and Viticulture. 55, 304A.
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of “Brett character”. Food Chemistry. 114, 15 - 19.
Shapiro, S. S. & Wilk, M. B. (1965). An analysis of variance test for normality (complete
samples). Biometrika, 52, 591-611.
Ugarte, P., Agosin, E., Bordeu, E. & Villalobos, J. I. (2005) Reduction of 4-ethylphenol and 4-
ethylguaiacol concentration in red wines using reverse osmosis and adsorption.
American Journal of Enology and Viticulture. 56, 30 – 36.
Wirz, D. O., Heymann, H. & Bisson, L. F. (2004) Descriptive analysis of Brettanomyces-infected
Cabernet Sauvignon wines. American Journal of Enology and Viticulture. 55, 303A.
85
Chapter 5: 6BInvestigation into the sensory effects and interactions of four Brett-related compounds in Pinotage red wine
1 INTRODUCTION............................................................................................................. 86
2 THEORY OF MULTIWAY METHODS............................................................................ 87
3 MATERIALS AND METHODS........................................................................................ 90
3.1 Central composite design.......................................................................................... 90
3.2 Wine samples.............................................................................................................. 91
3.3 Chemicals and spiking............................................................................................... 91
3.4 Profiling of central composite design combination samples ................................ 93
3.5 Consumer analysis..................................................................................................... 95
3.6 Data analysis............................................................................................................... 96
3.6.1 Central composite design...................................................................................... 96
3.6.2 Consumer panel...................................................................................................... 97
4 RESULTS AND DISCUSSION ....................................................................................... 97
4.1 Profiling of samples ................................................................................................... 97
4.1.1 Berry-like ................................................................................................................. 97
4.1.2 Sick-sweet ............................................................................................................. 100
4.1.3 Elastoplast™ ......................................................................................................... 104
4.1.4 Medicinal ............................................................................................................... 109
4.1.5 Smoky/Savoury..................................................................................................... 111
4.1.6 Pungent ................................................................................................................. 112
4.1.7 Overall effects using different methods of multivariate analysis .................... 114
4.2 Consumer analysis................................................................................................... 124
5 CONCLUSIONS............................................................................................................ 130
6 REFERENCES.............................................................................................................. 132
86
1 INTRODUCTION
Since its discovery as wine spoilage microorganism, Brettanomyces has been the subject of
many microbiological investigations. However, although the spoilage of wine by Brettanomyces
mainly affects its sensory properties, the organoleptic effects of Brettanomyces spoilage are still
relatively poorly understood. 4-ethylphenol and 4-ethylguaiacol are generally accepted to be the
most important Brettanomyces related spoilage compounds, but recently attention has been
drawn to isovaleric acid and 4-ethylcatechol (Licker et al., 1999; Fugelsang & Zoecklien, 2003;
Hesford & Schneider, 2004; Hesford et al., 2004; Larcher et al., 2008; Romano et al., 2009).
Isovaleric acid was pointed out by Licker et al. (1999) as one of the most odour-active
substances with regard to Brett character. However, other authors (Fugelsang & Zoecklien,
2003) found no correlation between the presence of isovaleric acid and sensory Brett character.
In a more recent study, Romano et al. (2009) found a significant correlation between isovaleric
acid levels and sensory Brett character. This correlation led them to conclude that isovaleric
acid is a marker for Brett character. They also found that the presence of isovaleric acid
increased the detection thresholds of 4-ethylphenol and 4-ethylguaiacol. Both Romano et al.
(2009) and Fugelsang and Zoecklein (2003) speculated about synergistic sensory effects
involving the volatile phenols and isovaleric acid.
In 2004, a third ethylphenol, namely 4-ethylcatechol, was linked to Brettanomyces
spoilage (Hesford & Schneider, 2004; Hesford et al., 2004). This compound is produced by the
same enzymatic pathway as 4-ethylphenol and 4-ethylguaiacol from caffeic acid as precursor
(Hesford & Schneider, 2004). Its relatively recent linkage to Brettanomyces (in contrast to 4-
ethylphenol and 4-ethylguaiacol) is ascribed to the fact that 4-ethylcatechol is less volatile than
4-ethylguaiacol and 4-ethylphenol, and therefore either requires a derivitisation step prior to GC-
MS analysis or HPLC analysis (Hesford & Schneider, 2004; Hesford et al., 2004, Larcher et al.,
2008). The sensory character of 4-ethylcatechol has been described as “horsey” (Hesford &
Schneider, 2004), and smoky (Larcher et al., 2008). It has, however, been found that 4-
ethylcatechol does not have as intense a sensory effect as the other volatile phenols, and it has
been speculated that its sensory effect is mainly due to synergism with the other Brett-related
compounds (Larcher et al., 2008; Curtin et al., 2008). A study has found that when 4-
ethylcatechol was present in combination with low levels of the other two ethylphenols, it
caused a smoky aroma, and that it suppressed Brett character when 4-ethylphenol and 4-
ethylguaiacol were at high levels (Curtin et al., 2008).
Although the presence of excessive levels of ethylphenols have long been considered to
be a wine spoilage defect (Eteiévant, 1981), and some authors (Eteiévant et al., 1989;
Chatonnet et al., 1992) have attempted to correlate specific levels of ethyl phenols to consumer
rejection of red wine, consumer analysis has only recently been performed on wines spoiled
with Brettanomyces (Lattey et al., 2007; Curtin et al., 2008). It was found that wines with Brett
87
character are indeed less-liked than unspoiled wine, regardless of the lack of consumer
knowledge of this defect. However, many people are of the opinion that a “little bit of Brett is
nice”, as its presence adds to the complexity of wine (Oelofse et al., 2009). Brett character is
also specifically associated with some French wines and is ascribed as being part of “terroir”.
The central composite design is an experimental design method that is used as a
screening design for interactions. It is specifically suited to situations where the number of
samples that can be analysed is limited (Esbensen, 2002), which is a common challenge
regarding sensory analysis. The central composite design has recently been used to investigate
synergistic effects between the odours of four lipid oxidation by-products (Venkateshwarlu et al.,
2004). Other sensory applications include wheat flour extrusions (Cheng & Fris, 2008) and
orange beverages (Mirhosseini et al., 2008; Mirhosseini et al., 2009).
Performing sensory analysis on complex mixtures is limited and influenced by several
factors. Firstly, there is a limit to the complexity of an odour that can accurately be profiled by
judges (Lawless, 1999; Hughson & Boakes, 2002), which can be further limited by the number
of compounds present (Livermore & Laing, 1996; Jinks, & Laing, 1999; Le Berre et al., 2008a.)
This situation is not improved by providing more information to judges (Jinks, & Laing, 1999).
Furthermore, certain combinations of compounds enhance one another, whereas other
combinations lead to suppressant effects (Laing, 1988; Laska & Hudson, 1991; Grosh, 2001;
Anatosova et al., 2005a; Anatosova et al., 2005b). In addition, compounds have also been
shown to interact differently at their detection thresholds compared to below their detection
thresholds (Anatosova et al., 2004). Blending effects also have a severe influence on the final
odour character of a mixture (Brodin et al., 2007; Le Berre et al., 2008b). Another factor
specifically influencing wine is that of semantic (linguistic) grouping; certain similar odours group
together and are difficult to differentiate from one another if they form part of the same semantic
group (Escudero et al., 2007). These groups include similar specific odours (for example,
blueberry, blackberry or strawberry) that all fall under a larger, more general descriptor like
fruity, woody and sweet. Furthermore, effects found in intensity modelling in simple mixtures
cannot be extrapolated to mixtures that are more complex (Brossard et al., 2007).
In light of the above, the aim of this study was to profile combinations of 4-ethylphenol,
4-ethylguaiacol, 4-ethylcatechol and isovaleric acid in a young Pinotage red wine. The
combinations were determined from a central composite design. The samples used for the
central composite design were also presented to consumers in order to determine the consumer
acceptance.
2 THEORY OF MULTIWAY METHODS
This section outlines the basic principles of the multiway method PARAllel FACtor analysis
(PARAFAC), which has been applied in this paper. The outline is included due to novelty of
88
applying PARAFAC to sensory data, as examples in literature are limited. For other examples of
the application of PARAFAC to sensory data, see Pravdova et al. (2002), Cocchi et al. (2006),
Masino et al. (2008), and Bro et al. (2008).
Sensory data occur in a data cube (a three-way data matrix) as it contains one data
table containing samples and attributes per judge. This can be seen in XFigure 5.1 X. This data
cube is usually simplified by unfolding the cube, and finding the means (Bro et al., 2008).
However, the use of multiway models allow for a simpler interpretation of this kind of data. In
this paper, PARAFAC will be used, but Tucker3 (a multiway method similar to PARAFAC) is
referred to in aid of the explanation of the difference between PCA and PARAFAC.
Sam
ples
Sa
mpl
es
Figure 5.1. Structure of sensory data sets. Sensory data occur in a cube (Samples × Attributes
× Judges) but is usually unfolded to form a two-way matrix.
XFigure 5.2X gives an outline of how PCA, Tucker3 and PARAFAC relate to one another.
Tucker3 is a constrained version of PCA, whereas PARAFAC is a constrained version of the
Tucker3 model. These constraints allow for the calculation of more robust and interpretable
models, by preventing the model to include and model as much of the noise. Due to the
constraint hierarchy, a two-way PCA model will always fit data better than a Tucker3 model,
which will always fit data better than a PARAFAC model. However, PARAFAC focuses on
modelling the systemic part of the data, because the final model has fewer parameters than a
PCA model. This has the inherent implication that PARAFAC models are more easily
interpreted than unfolded PCA models (Bro, 1997).
Apart from the fact that PCA is performed to a two-dimensional data matrix and
PARAFAC is performed on a three-dimensional data array, the largest difference between PCA
and PARAFAC is the way in which principal components or factors are calculated. In PCA, the
components are calculated separately, and changing the number of components does not affect
the model. The first principal component lies in the direction of the most variance, and
subsequent orthogonal components are calculated to model the remainder of the variance.
89
X
CK
F
ZF
F
F
AI
FX
K
I
J
B
J
F
E
J
I
K
X
K
I
J AI
LE
K
I
J
B
J
MZ
L
M
N
I
J × K
=
=
=
N
CK
+
+
E
AI
+
J × K
I
1) PCA
2) Tucker3
3) PARAFAC
B
J K
Figure 5.2. Graphical representation of unfolded PCA, Tucker3 and PARAFAC: 1) Unfolded
PCA produces a scores matrix and a loading matrix (A and B) and a two-dimensional error
matrix (E). 2) Tucker3 produces three loading matrices (A, B, and C), with a variable numbers
of factors in each matrix (L, M and N). This results in a core array (Z) which can have variable
proportions, corresponding to the number of factors in each loading matrix (L, M and N) and a
three-dimensional error array (E). 3) PARAFAC also produces three loading matrices (A, B and
C), which all contain the same number of factors (F). The resulting core array (Z) has equal
proportions (F) in all three dimensions, and the error array (E) is three-dimensional.
90
The factors in PARAFAC are not constrained due to orthogonality, and are calculated
simultaneously. These factors are calculated so that they collectively model the maximum
amount of variance in the data. This is done by estimating two factors, calculating the remaining
factors by least squares regression and recalculating the factors until convergence occurs.
Convergence occurs when recalculating the factors no longer changes the model. This means
that the number of factors in the model significantly affects the model itself, making the choice of
number of factors of utmost importance (Bro, 1997).
Tucker3 differs from PARAFAC in that Tucker3 can have different numbers of factors for
each of the three modes (sets of results), whereas PARAFAC has the constraint of having the
same number of all factors for all modes. In the Tucker3 model, the core matrix, Z, (see XFigure
5.2 X) is used to interpret these different loading matrices. The core array gives information about
which interactions are significant, the magnitude of their significance and the nature of the
interactions (positive or negative). A further constraint of the PARAFAC model is that the super
diagonal elements of the core element have to be equal to 1 (or very close to 1). This means
that the interactions between all the loadings vectors have to be of the same importance and in
the same direction. This has the implication that PARAFAC can be interpreted in a similar
manner to PCA. The core consistency test is a test evaluating whether these elements are in
fact equal to 1, and can be used to evaluate whether the appropriate number of factors have
been selected for a PARAFAC model (Pravdova et al., 2002).
3 MATERIALS AND METHODS
3.1 31BCentral composite design
The current experimental procedure was conducted with a central composite design
(Venkateshwarlu et al., 2004). This design consists of five levels of up to six components, and
can be used to test interactions between different factors. However, it differs from a factorial
design as not all combinations of all factors are tested. It is recommended as an appropriate
design where there is a limit to the number of experiments that can be performed, making it
particularly suitable for sensory experiments (Esbensen, 2002). If n is the number of factors to
be tested, the design has n2 “cube points” and 2n “star points”, as well as a “centre point”, which
is replicated. The simplest variation of the design, a two-factor central composite, is shown in
XFigure 5.3X.
This design method was used in the current study to investigate the interaction between
four different Brettanomyces related components. This gave rise to a design with eight star
points, 16 cube points and one centre point, and therefore 25 samples in total. The
compositions of the different samples will be discussed in the appropriate section.
91
Figure 5.3. General layout of a two-factor central composite design. a) designates a cube point,
b) designates a star point and c) designates the centre point.
3.2 32BWine samples
Two hundred litres of Pinotage was locally obtained from wine producer (Distell Group Ltd,
Stellenbosch, South Africa) during the course of 2009. This wine had been made using
standard red wine making practices and completely underwent both alcoholic and malolactic
fermentation and was bottled manually at the Department of Viticulture and Oenology,
Stellenbosch University, South Africa. The wine used had an alcohol content of 11.6 %, and a
pH of 3.65. This wine is identical to the one used in Chapter 4.
After bottling, the wine was analysed to determine the levels of 4-ethylphenol, 4-
ethylguaiacol, 4-ethylcatechol and isovaleric acid present. The analyses for 4-ethylphenol, 4-
ethylguaiacol and isovaleric acid were performed using GC-MS using a DB-FFAFF(60m x 320.0
µm x 0.5 µm) column, and the levels of 4-ethylcatechol were determined using HPLC-MS/MS
using a C18 (2.1 x 50 mm with guard) column. All these analyses were conducted by an
accredited laboratory (Quantum Laboratories, South Africa).
It was found that the base wine contained less than 10 µg/L all three volatile phenols
investigated (4-ethylphenol, 4-ethylguaiacol and 4-ethylcatechol). The wine was therefore
considered not to be spoiled by Brettanomyces. Furthermore, the wine contained 355 µg/L
isovaleric acid. This value is in agreement with minimum values of this compound found in red
wines (Francis & Newton, 2005).
3.3 33BChemicals and spiking
Solutions of 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid (Aldrich, South
Africa) were prepared in 99.5% ethanol (Merck Chemicals, South Africa). Four solutions of each
a
bc
92
compound was made, namely of 10 mg/mL, 1 mg/mL, 100 µg/mL and 10 µg/mL. These
solutions were used to produce wine samples spiked with the exact concentrations of the latter
compounds. The five levels used in this study were based on the central composite design, and
were determined by the methodology outlined below.
The concentrations of the four different compounds to be used during this study were
predetermined as follows: Firstly, detection thresholds were determined as described in Chapter
3. This detection level was set be approximately level 2 (the lower cube level, XFigure 5.3 X) in the
central composite design. This decision was made to allow for the investigation of sensory
effects and interactions of the four compounds at, below, and above their detection thresholds.
The highest level (level 5) corresponded to the highest level that is likely to occur naturally in
wine. This level was determined with the guidance of literature, as well as with the help of wine-
industry experts (Thales, South Africa). These levels were also subjected to extensive sensory
pre-screening, to ensure that, in the wine used for this study, the concentrations conform to the
sensory criteria set in t. These two levels were used to calculate the rest of the levels of the
central composite design.
The central composite design levels were chosen along the same logarithmic scale as
the eight levels used for the detection thresholds. These levels all satisfied the equation c = ab n-
1 , where c is the concentration, a is the starting point (level 1), b is the multiplication factor and
n is the level number. Levels chosen in the design were therefore on a simple numerical scale
that corresponded to that used in the detection thresholds. This was done to allow to
compensate for the fact that the distances between levels in the central composite design are
predetermined. This created a difficulty as the highest levels that this research wished to
investigate were between five and twenty times the magnitude of the detection threshold. The
use of a logarithmic scale compensated for this difference in magnitude, and allowed for the
concentrations in question to be fitted to the central composite design. For this reason, two sets
of levels were used in this study. The “design” levels were those determined by the design, and
fall between 0 and 13, and are shown in XTable 5.2X. The actual concentrations used were
determined using the equation c = ab n-1 and are shown in XTable 5.1X. In both these tables, levels
are numbered 1 to 5, and these values are used throughout this chapter when referring to
samples, with reference to the concentrations and design levels where necessary.
Table 5.1. Actual concentrations of Brett-related compounds tested (µg/L).
Level
Compound 1 2 3 4 5
4-ethylphenol 82 227 623 1711 4695
4-ethylguaiacol 65 117 230 381 688
4-ethylcatechol 181 290 465 745 1193
Isovaleric acid 381 431 577 997 2210
93
The final concentration ranges were confirmed with the help of wine-industry experts
(Thales, South Africa) by means of several sessions of consensus sensory analysis. Before
commencement of the sensory tests, the final samples were also subjected to sensory pre-
assessment.
Table 5.2. Design levels for spiking of wines with Brett-related compounds (arbitrary scale).
Level Compound
1 2 3 4 5
4-ethylphenol 0.5 3.5 6.5 9.5 12.5
4-ethylguaiacol 0.75 2.5 4.5 6 7.75
4-ethylcatechol 2.8 4.2 5.6 7 8.4
Isovaleric acid 2.5 4.5 6.5 8.5 10.5
The final sample set is shown in XTable 5.3 X. Please note that throughout this document,
samples will be referred to by means of four-digit codes. These codes correspond to the levels
used in XTable 5.1X and XTable 5.3X, and are listed in the order 4-ethylphenol, 4-ethylguaiacol, 4-
ethylcatechol, isovaleric acid. For example, sample 2442 contains level 2 (227 µg/L) of 4-
ethylphenol, level 4 (381 µg/L) of 4-ethylguaiacol, level 4 (745 µg/L) of 4-ethylcatechol and level
2 (431 µg/L) of isovaleric acid.
3.4 34BProfiling of central composite design combination samples
The combination samples, shown in XTable 5.3X, were profiled after the descriptive analysis of the
singular compounds had been completed. The sensory panel used consisted of 10 judges, of
whom seven participated in a study involving the profiling of the compounds in this study, and
therefore had experience in analysis for Brett character. These judges were therefore already
familiar with the aroma profiles associated with the compounds used in this study.
Training commenced with three sessions of general training, which allowed the judges to
familiarise themselves with the samples in question. These samples were presented in three
different subsets, which were analysed on three different days. The first subset was the eight
star samples, the second subset the first eight cube samples (cube 1 – cube 8), and the third
subset the final 8 cube samples (cube 9 – cube 16). During all three training sessions, the
reference standards listed in XTable 5.4 X were made available to the judges. Additionally, the
centre sample, as well as the highest levels of the four individual compounds (level 5) was
presented to the judges to serve as reference standards. During training, all samples were
labelled using a system that indicated their composition to the judges.
94
Table 5.3. Compositions of the different samples in the design.
The general training was followed with three subsequent training sessions. During each
of these sessions, one of the three sets used during general training was presented to the
judges, and the samples were scaled on a 100 mm unstructured line scale according to the
attributes in XTable 5.4X. The reference standards listed in XTable 5.4X were adapted from Noble et
al. (1987) and used during a previous study (Chapter 4). Note that the descriptors used in this
study differ slightly from those used in Chapter 4. The number of descriptors were reduced as
the panel could not accurately perceive all the descriptors used in the samples containing the
individual compounds in the combination samples. An example of such a descriptor is the
leather like attribute used in Chapter 4. In the samples in Chapter (4-ethylphenol) where this
descriptor was relevant, this descriptor could be distinguished by the panel from the
Elastoplast™ descriptor. However, no such distinction could be made in the combination
samples.
The final profiling analyses were conducted by 10 trained assessors in booths with
standard artificial daylight lighting and temperature control at 20°C ±1°C. The wine was
analysed in standard ISO wine tasting glasses, sample size was 20 mL and samples were
Compound
Sample 4-ethylphenol 4-ethylguaiacol 4-ethylcatechol Isovaleric acid
Centre 3 3 3 3
Star 1 5 3 3 3
Star 2 1 3 3 3
Star 3 3 5 3 3
Star 4 3 1 3 3
Star 5 3 3 5 3
Star 6 3 3 1 3
Star 7 3 3 3 5
Star 8 3 3 3 1
Cube 1 4 4 4 4
Cube 2 4 4 4 2
Cube 3 4 4 2 4
Cube 4 4 4 2 2
Cube 5 4 2 4 4
Cube 6 4 2 4 2
Cube 7 4 2 2 4
Cube 8 4 2 2 2
Cube 9 2 4 4 4
Cube 10 2 4 4 2
Cube 11 2 4 2 4
Cube 12 2 4 2 2
Cube 13 2 2 4 4
Cube 14 2 2 4 2
Cube 15 2 2 2 4
Cube 16 2 2 2 1
95
served at 20°C ±1°C. Prior to the analysis, the wine glasses containing the wine samples were
covered with plastic lids. This prevented the aroma of the wine from escaping or contaminating
the laboratory environment.
The analyses were performed as six separate tests with two replications. During each
test, the judges received five samples for analysis, which included the centre sample and four
other samples from the design. The samples for analysis were labelled with a random three-digit
code and were presented in a randomised order. The judges also received the non-spiked
control wine, as well as a wine sample containing the highest concentration (level 5) of each
compound. The first two tests contained the star samples, and the last four tests contained the
cube samples. This division was selected based on the recommendation by Esbensen (2002).
However, the samples were randomised between tests and judges.
Table 5.4. Descriptors and reference standards used during descriptive analysis of wines
spiked with a combination of four Brett-related compounds. Note that two reference standards
were presented for the smoky-savoury attribute and that the panel was instructed to form a
cognitive combination between these two attributes.
Descriptor Definition Reference standard
Berry-like Typical wine-sweet berry-like aroma Control sample
Sick-sweet Atypical wine-sweet – sweet smell that is uncharacteristic/ unnatural for wine (confected)
None
Elastoplast™ Smell associated with Elastoplast™ or Band-Aid™
1 piece of Elastoplast™ fabric bandage
Medicinal (Listerine) Minty smell associated with mouthwash Wine spiked with Listerine™ (3 drops in 100 mL wine)
Smoky-savoury The smell associated with smoked food A cognitive combination between smoke essence (3 drops in 100 mL wine) and soy sauce (1 ml in 100 mL wine)
Pungent Sweaty/ rancid/ cheesy/ vinegary A small piece of mild blue cheese (Simonzola™)
3.5 35BConsumer analysis
A consumer analysis was performed using 100 consumers sourced from the Stellenbosch area.
Fifty percent of the consumers indicated that they enjoyed drinking wine but did not know much
about it, whereas the remaining fifty percent indicated that they knew a fair amount about wine.
The consumer analysis was performed using an incomplete block design; where each
consumer received a set of five of the spiked samples and an unspiked (control) sample. All the
samples were labelled with a random three-digit code and were presented in a random order.
Questions were also posed about the wine consumption patterns of the consumers, as well as
96
their degree of wine expertise. Biographical data (gender, age) were obtained for each
individual consumer.
The consumers were instructed to smell and taste the wine samples in the order
presented. Each consumer received a water biscuit (Carr, UK) and water to clean their palate
before and after tasting each sample. The consumers had to indicate their degree of liking of
the samples on a standard nine-point hedonic scale where 1 represents Dislike extremely and 9
represents Like extremely (Lawless & Heymann, 1998). In order to prevent the interference
effect of information regarding wine quality on hedonic ratings (Sigrist & Cousin, 2009), no
information was provided regarding the purpose of test.
3.6 36BData analysis
3.6.1 50BCentral composite design
The data were analysed using SAS® software (Version 9; SAS Institute Inc, Cary, USA) and
subjected to the Shapiro-Wilk test for non-normality of the residuals (Shapiro & Wilk, 1965). If
non-normality was found to be significant (P≤0.05) and caused by skewness, the outliers were
identified and removed until the data were normal or symmetrically distributed (Glass et al.,
1972). Using line plots indicating temporal stability and internal consistency, single odd judges
were identified and removed. The final analysis of variance (ANOVA) was performed after all
the above-mentioned procedures have taken place. Student’s t-least significant difference
(LSD) was calculated at the 5% significance level to compare treatment means.
If second–order interactions between compounds for specific descriptors were found to
be significant, a curve was fitted between the means of these compounds using the TableCurve
3D software (Version 4.0, SYSTAT Software Inc). If third-order interactions were found to be
significant, the ANOVA was repeated with one of the compounds in the interaction kept as a
constant. The results of these analyses were interpreted in a similar manner to those of the
original analyses.
PCA was performed on the mean values of the overall profiles of the wines.
Furthermore, Partial Least Squares regression (PLS) was performed using the chemical
information in the X-space and the sensory descriptive data in the Y-space. During this analysis,
the chemical data was converted to category variables, in order to investigate the difference in
effect of the different levels of the compounds. Both these multivariate techniques were
performed using the XLStat software package (Version 2009.5.0.1, Addinsoft, SARL, Paris,
France).
97
PARAFAC was performed in Matlab (Mathworks, Inc) using the PLS toolbox version 5.3
(EigenVector Research Inc, Manson, WA, US). PARAFAC was performed on the same dataset
as the ANOVA, i.e. the raw scores produced by judges tested for outliers.
3.6.2 51BConsumer panel
The unspiked sample was considered as a standard sample for all the consumers and therefore
the data were pooled for analysis of variance (ANOVA) (SAS®, Version 9; SAS® Institute Inc,
Cary, USA.). The Shapiro-Wilk test was used to test for non-normality of the residuals (Shapiro
& Wilk, 1965). If skewness appeared to be the result of outliers these outliers were identified
and discarded until the data were considered normal or symmetrically distributed (Glass et al.,
1972). Furthermore, two types of preference mapping were performed. Firstly, external
preference mapping was performed by superimposing the consumer data on the PCA map of
the sensory profiles. Secondly, PLS was performed with the sensory data in the X-space and
the consumer data in the Y-space. Both these techniques were performed using the software
package XLStat (Version 2009.5.0.1, Addinsoft, SARL, Paris, France).
4 RESULTS AND DISCUSSION
4.1 37BProfiling of samples
Please note that in this section, occasional reference is made to results obtained during singular
profiling of these four compounds. These results are discussed in Chapter 4, and are not
included in this section for the sake of simplicity.
4.1.1 52BBerry-like
The ANOVA for the berry-like attribute is shown in XTable 5.5 X. As can be seen in this table, a
significant interaction (p = 0.0357) was found between the level of 4-ethylphenol and the level of
4-ethylcatechol. This interaction was plotted and a curve with the equation z = 233.89 +
0.159lnx + 0.116 (lnx)2 + 0.412(lnx)3 + 183.96y + 54.03y2 + 6.70y3 + 0.29y4 was obtained. This
function has a coefficient of determination of R2= 0.52 and is shown in XFigure 5.4X. This is not
considered a particularly good fit, but in spite of this, several general trends can be observed
from the curve.
98
Table 5.5. ANOVA performed on sensory data obtained for the berry-like attribute.
Source DF Mean Square Pr > F
4-EP 4 797 0.0001
4-EG 3 228 0.033
4-EC 3 59.9 0.51
ISOV 3 338 0.005
4-EP*4-EG 1 0.58 0.93
4-EP*4-EC 1 344 0.036
4-EG*4-EC 1 114 0.22
4-EP*ISOV 1 0.64 0.93
4-EG*ISOV 1 207 0.10
4-EC*ISOV 1 72.8 0.33
4-EP*4-EG*4-EC 1 76.5 0.32
4-EP*4-EG*ISOV 1 0.021 0.99
4-EP*4-EC*ISOV 1 4.75 0.80
4-EG*4-EC*ISOV 1 8.34 0.74
4-EP*4-EG*4-EC*ISOV 1 119 0.22
Error 495 77.6
12.5
10
7.5
52.5
C_EP
98
76
54
3
C_EC
2.5 2.5
5 5
7.5 7.5
10 10
12.5 12.5
15 15
17.5 17.5
Ber
ry_l
ike
Ber
ry_l
ike
Figure 5.4. Interaction between 4-ethylphenol and 4-ethylcatechol in terms of berry-like
character in Pinotage red wine. Note that the figure shows the highest values in the front, in
order for the curve to be visible. C_EC refers to the “design” level of 4-ethylcatechol, and C_EP
refers to that of 4-ethylphenol.
The berry-like attribute firstly generally decreased with an increase in 4-ethylphenol, as
the curve slopes upwards in the direction of decreasing 4-ethylphenol concentration. 4-
ethylcatechol, however, appears to have a “wavy” effect associated with it, in other words –
99
higher berry-like character at odd levels than at even levels, with the highest berry-like character
found at the highest and lowest levels. At first, this may appear to be strange, but this pattern
could be explained by considering the design.
At the highest and lowest levels of 4-ethylcatechol, all other compounds are at level 3,
and the berry-like suppressant effect is the least. However, when 4-ethylcatechol is at level 3
(design level 5.6), it is combined with the lowest or highest levels of the other compounds, which
all have been shown to a have a suppressant effect on berry-like character. In addition, when 4-
ethylcatechol is at levels 2 (design level 4.2) and 4 (design level 7), it is combined with several
other levels of all the other compounds, some combinations containing all the compounds at
their highest levels, which may have the highest suppressant effect on the berry-like attribute.
Therefore, it can be concluded that 4-ethylcatechol had very little suppressant effect on the
berry-like character, and that its observed interaction with 4-ethylphenol is most probably an
artefact of the central composite design.
As is also evident from XTable 5.5 X, the effect of isovaleric acid can be interpreted as the
only significant main effect (p = 0.0048). The effect of an increase in isovaleric acid is shown in
XFigure 5.5 X and XTable 5.6 X. As can be seen from both this table and this figure, there is a general
decrease in the level of berry-like character with an increase in concentration of isovaleric acid.
Level 1 (381 µg/L) and level 3 (577 µg/L) are significantly different, and level 3 and level 5 (2210
µg/L) are also significantly different. However, there is no significant difference between level 2
(577 µg/L) and 4 (997 µg/L) and either level 3 (577 µg/L) or 5 (2210 µg/L). This is probably
because when isovaleric acid is at levels 2 and 4 the other compounds are all either at level 2 or
4. This means that the lower values for berry-like character observed at level 2 and 4 of
isovaleric acid may in fact be due to the combined suppression of berry-like character by the
higher levels (level 4) of the other compounds. However, the overall trend of a drop in berry-like
character is similar to that observed in all the compounds when profiled on their own. An
important difference, however, is the fact that the suppression occurs over a different range
than when isovaleric acid in profiled in its singular state. The suppression of berry-like character
occurs from 34 mm on the 100 mm scale at level 1 to 26 mm on the 100 mm scale at level 5
when only isovaleric acid is added to the wine but from 16 mm at level 1 to 6 mm at level 5
when in combination with the other compounds. This further indicates that there is a synergistic
effect in the suppression of berry-like character between the different compounds.
Although the suppression of berry-like character by 4-ethylguaiacol is evident in Chapter
4, no significant effect or significant interactions were observed in XTable 5.5 X, and the compound
is therefore not discussed.
100
0
2
4
6
8
10
12
14
16
18
1 2 3 4 5
Level of isovaleric acid
Inte
ns
ity
of
Ber
ry-l
ike
attr
ibu
te
Isovalericacid
Figure 5.5. Effect of isovaleric acid on berry-like character during combination profiling in
Pinotage red wine.
Table 5.6. Least significant difference groups of isovaleric acid on berry-like character profiling
in Pinotage red wine.
Level Mean1
1 16.7 a
2 8.8 b c
3 11.9 b
4 8.8 b c
5 6.2 c
Least Significant Difference (p = 0.05) 3.94 1 Values with the same superscript are not significantly different
4.1.2 53BSick-sweet
The results of the ANOVA for the sick-sweet characteristic are shown in XTable 5.7 X. As shown in
this table, a significant interaction (p = 0.0302) was found between 4-ethylcatechol and 4-
ethylphenol. Although this interaction was significant, no curve could be fitted that had an R2
value of greater than 0.5. For this reason, this interaction was investigated manually. The effect
of 4-ethylphenol on 4-ethylcatechol is shown in XFigure 5.6X, and the effect of 4-ethylcatechol on
4-ethylphenol is shown in XFigure 5.7 X. The effects are summarised in XTable 5.8 X. In XTable 5.8 X, an
overall change in sick-sweet of less than one unit was taken to be no change.
As can be seen in XFigure 5.6 X, XFigure 5.7 X and XTable 5.8 X, both these compounds
prevented an increase in sick-sweet character by the other compound when they were at level 2
or level 3 (therefore at detection threshold or just above detection threshold). This means that
both these compounds interfered with the sick-sweet effect of each other at these levels. At
101
level 4, both compounds allowed a change in sick-sweet character to occur with an increase in
the other compound. However, none of these increases were severe enough that the effect
could be described as synergistic.
Table 5.7. ANOVA performed on the sensory data obtained for the sick-sweet attribute.
Source DF Mean Square Pr>F
4-EP 4 73.1 0.051
4-EG 3 101 0.021
4-EC 3 26.6 0.45
ISOV 3 5.53 0.91
4-EP*4-EG 1 22.8 0.38
4-EP*4-EC 1 145 0.030
4-EG*4-EC 1 1.23 0.84
4-EP*ISOV 1 0.455 0.90
4-EG*ISOV 1 0.451 0.90
4-EC*ISOV 1 28.4 0.34
4-EP*4-EG*4-EC 1 19.4 0.43
4-EP*4-EG*ISOV 1 4.15 0.71
4-EP*4-EC*ISOV 1 0.442 0.91
4-EG*4-EC*ISOV 1 65.6 0.15
4-EP*4-EG*4-EC*ISOV 1 13.2 0.51
Error 493 30.7
0
1
2
3
4
5
6
7
1 2 3 4 5
Level of 4-ethylcatechol
Inte
nsi
ty o
f S
ick-
swee
t a
ttri
bu
te
EP=2
EP=3
EP=4
Figure 5.6. Effect of different levels of 4-ethylphenol on the sick-sweet character caused by 4-
ethylcatechol in Pinotage red wine.
102
0
1
2
3
4
5
6
7
1 2 3 4 5
Level of 4-ethylphenol
In
ten
sit
y o
f S
ick
-sw
eet
att
rib
ute
EC=2
EC=3
EC=4
Figure 5.7. Effect of different levels of 4-ethylcatechol on the sick-sweet character caused by 4-
ethylphenol in Pinotage red wine.
Table 5.8. Summary of the effects of 4-ethylphenol and 4-ethylcatechol in combination on sick-
sweet character in Pinotage red wine.
Compound Level Effect Expecteda
2 Sick-sweet stays constant No
3 Sick-sweet stays constant No 4-ethylphenol
4 Sick-sweet increases with an increase with 4-ethylcatechol Yes
2 Sick-sweet stays constant with increase of 4-ethylphenol No
3 Sick-sweet stays constant No 4-ethylcatechol
4 Sick-sweet increases with an increase with 4-ethylphenol Yes a Expected according to previous results (Chapter 4)
Table 5.9. Change in the sick-sweet character with level of 4-ethylguaiacol in Pinotage red
wine.
Level Mean1
1 2.6 b
2 3.8 a b
3 4.1 a b
4 5.7 a
5 3.3 a b
Least Significant Difference (p = 0.05) 2.48 1 Values with the same superscript are not significantly different
103
0
1
2
3
4
5
6
1 2 3 4 5
Level of 4-ethylguaiacol
Inte
nsi
ty o
f S
ick-
swee
t a
ttri
bu
te
4-EG
Figure 5.8. Effect of 4-ethylguaiacol on sick-sweet character in Pinotage red wine in
combination samples.
The only main statistical effect that could be interpreted was that of 4-ethylguaiacol
( XTable 5.7 X). The change in sick-sweet character with the increase in 4-ethylguaiacol is shown in
XTable 5.9X and XFigure 5.8
XFigure 5.8 X and XTable 5.9 X show an unexpected trend in terms of sick-sweet character.
The means for levels 1 and 4 differed significantly from one another, but none of the other
means differed significantly from either level 1 or level 4. This may be because the greatest
synergistic effect in terms of sick-sweet happens at level 4 of 4-ethylguaiacol. It is, however,
interesting to note that the sample with the highest mean for sick-sweet was the one that
contained level 4 of all four the compounds. This mean is 7.72, which was more than one unit
higher than its closest successor. It could be that the synergistic effect between the different
compounds in terms of sick-sweet character has caused this.
It is finally interesting to note that the overall increase in sick-sweet character was
extremely small – from 2 to about 6, which may be due to several reasons. Firstly, the sick-
sweet characteristic is one that the panel found particularly difficult to define, and some panel
members had difficulty picking up this characteristic in the combination samples. This is not
unlike what has been reported in literature: that the performance of panel members decrease
with complexity of samples (Hughson & Boakes, 2002). For this reason, this difficult-to- perceive
characteristic could not be picked up above the other aromas being profiled. Another possibility
is the fact that the sick-sweet attribute (which becomes perceptible due to the suppression of
the natural berry-like character in the wine) is not perceptible in samples that contain a
combination of these Brett-related compounds (and therefore in wines spoiled by
Brettanomyces). This is particularly plausible, as, to date, the term “sick-sweet” has not been
coupled with wines naturally spoiled by Brettanomyces. However, the descriptor was included in
104
this study as it was used in Chapter 4, and was distinctly perceived in the samples that
contained the individual compounds.
4.1.3 54BElastoplast™
The ANOVA for the Elastoplast™ attribute, as seen in XTable 5.10 X, showed two significant three-
way interactions: firstly between 4-ethylphenol, 4-ethylguaiacol and 4-ethylcatechol (p =
0.0062), and secondly between 4-ethylphenol, 4-ethylguaiacol and isovaleric acid (p = 0.0253).
This indicates that 4-ethylphenol interacts significantly with three the other three compounds in
term of the Elastoplast™ descriptor. As these three-way interactions cannot be directly
interpreted, ANOVA’s were performed at constant levels of 4-ethylcatechol and isovaleric acid in
order to investigate the interaction between 4-ethylphenol and 4-ethylguaiacol. Results for these
analyses are shown in XTable 5.11X, XTable 5.12X, XTable 5.13X and XTable 5.14 X.
Table 5.10. ANOVA performed on the sensory data obtained for the Elastoplast™ attribute.
Source DF Mean Square Pr>F
4-EP 4 4502 0.0001
4-EG 3 72.6 0.39
4-EC 3 131 0.14
ISOV 3 15.6 0.89
4-EP*4-EG 1 272 0.052
4-EP*4-EC 1 171 0.12
4-EG*4-EC 1 94.2 0.25
4-EP*ISOV 1 50.0 0.41
4-EG*ISOV 1 427 0.015
4-EC*ISOV 1 209 0.089
4-EP*4-EG*4-EC 1 544 0.006
4-EP*4-EG*ISOV 1 363 0.025
4-EP*4-EC*ISOV 1 31.2 0.51
4-EG*4-EC*ISOV 1 12.3 0.68
4-EP*4-EG*4-EC*ISOV 1 10.4 0.71
Error 491 72.1
Table 5.11. ANOVA of 4-ethylphenol and 4-ethylguaiacol at level 2 of 4-ethylcatechol in
Pinotage red wine.
Source DF Mean Square Pr>F
4-EP 1 787 0.0009
4-EG 1 95.7 0.24
4-EP*4-EG 1 23.4 0.56
Error 121 68.4
105
Table 5.12. ANOVA of 4-ethylphenol and 4-ethylguaiacol at level 4 of 4-ethylcatechol in
Pinotage red wine.
Source DF Mean Square Pr>F
4-EP 1 2158 0.0001
4-EG 1 17.3 0.65
4-EP*4-EG 1 809 0.002
Error 122 81.5
Table 5.13. ANOVA of 4-ethylphenol and 4-ethylguaiacol at level 2 of isovaleric acid in
Pinotage.
Source DF Mean Square Pr>F
4-EP 1 1055 0.0004
4-EG 1 301 0.052
4-EP*4-EG 1 3.11 0.84
Error 122 78.3
Table 5.14. ANOVA of 4-ethylphenol and 4-ethylguaiacol at level 4 of isovaleric acid in
Pinotage.
Source DF Mean Square Pr>F
4-EP 1 1753 0.0001
4-EG 1 134 0.19
4-EP*4-EG 1 656 0.004
Error 121 76.5
It was found that there was no significant interaction between 4-ethylphenol and 4-
ethylguaiacol at level 2 of either 4-ethylcatechol or isovaleric acid (see XTable 5.11 X and XTable
5.13 X). However, there were significant interactions at level 4 of 4-ethylcatechol (p = 0.0020) (see
XTable 5.12X) and at level 4 of isovaleric acid (p = 0.0041) (see XTable 5.14 X), which are shown in
XFigure 5.9X and XFigure 5.10 X.
The graph in XFigure 5.9X was fitted to the an equation of z = 20.09 - 4.222 lnx -112.33 /y +
0.1364 (lnx)2 + 77.52 /y2+ 40.82 (lnx)/y (R2 value of 0.8126). It can be seen that at level 2 of 4-
ethylcatechol, 4-ethylguaiacol appears to enhance the Elastoplast™ effect associated with 4-
ethylphenol with an increase in concentration. However, at level 4, 4-ethylguaiacol appears to
suppress the Elastoplast™ descriptor with an increase in concentration. This effect may be due
to a third-order interaction occurring at this level. XFigure 5.10 X (z = 2,61 + 1,03x – 15,46/y +
0,1200x2 +26,81/y2 +6,03 x/y, R2 = 0.8433) shows a similar overall effect. Both these effects
justify doing more two-way ANOVA’s at constant levels of 4-ethylguaiacol and therefore
investigating the interactions of 4-ethylcatechol and isovaleric acid with 4-ethylphenol. The
results of these ANOVA’s are shown in XTable 5.15 X, XTable 5.16 X, XTable 5.17 X and XTable 5.18 X.
106
34
56
78
9
C_EP
22.533.5
44.555.5
C_EG
0 0
2.5 2.5
5 5
7.5 7.5
10 10
12.5 12.5
15 15
17.5 17.5
Ela
stop
last
Ela
stop
last
Figure 5.9. Effect of 4-ethylphenol and 4-ethylguaiacol on the Elastoplast™ descriptor at level 4
of 4-ethylcatechol. C_EG designates the design level of 4-ethylguaiacol, and C_EP indicates
the design level of 4-ethylphenol.
34
56
78
9
C_EP
22.533.5
44.555.5
C_EG
0 0
2.5 2.5
5 5
7.5 7.5
10 10
12.5 12.5
15 15
17.5 17.5
Ela
stop
last
Ela
stop
last
Figure 5.10. Interaction between 4-ethylguaiacol and 4-ethylphenol on the Elastoplast™
descriptor at isovaleric acid level 4. C_EG designates the design level of 4-ethylguaiacol, and
C_EP indicates the design level of 4-ethylphenol.
107
Table 5.15. ANOVA on the effect of 4-ethylphenol and 4-ethylcatechol on Elastoplast™ when 4-
ethylguaiacol is at level 2.
Source DF Mean Square Pr>F
4-EP 1 2369 <.0001
4-EC 1 411 0.024
4-EP*4-EC 1 657 0.005
Error 122 79.0
Table 5.16. ANOVA on the effect of 4-ethylphenol and 4-ethylcatechol on Elastoplast™ when 4-
ethylguaiacol is at level 4.
Source DF Mean Square Pr>F
4-EP 1 622 0.0042
4-EC 1 42.9 0.44
4-EP*4-EC 1 47.1 0.42
Error 121 73.0
Table 5.17. ANOVA on the effect of 4-ethylphenol and isovaleric acid on Elastoplast™ when 4-
ethylguaiacol is at level 2.
Source DF Mean Square Pr>F
4-EP 1 2370 0.0001
ISOV 1 102 0.27
4-EP*ISOV 1 325 0.052
Error 122 84.3
Table 5.18. ANOVA on the effect of 4-ethylphenol and isovaleric acid on Elastoplast™ when 4-
ethylguaiacol is at level 4.
Source DF Mean Square Pr>F
4-EP 1 622 0.0036
ISOV 1 328 0.032
4-EP*ISOV 1 76.3 0.30
Error 121 70.4
When ANOVA’s were performed at constant levels of 4-ethylguaiacol, it was found that
there was a significant interaction (p = 0.0518) between 4-ethylphenol and isovaleric acid at 4-
ethylguaiacol level 2 (see XTable 5.17 X), as well as between 4-ethylphenol and 4-ethylcatechol (p
= 0.0046) (see XTable 5.15 X). These interactions are shown in XFigure 5.11 X (Z = 1.25 + 0.135x +
45.7/y + 0.31x2 + 42.8/y2 + 15.1 x/y, R2= 0.87) and XFigure 5.12 X (Z = 10.76 – 2.07lnx – 11.66lny -
2.09(lnx)2 – 0.69(lny)2 + 9.68lnxlny, R2 = 0.71.)
108
34
56
78
9
C_EP
44.5
55.5
66.5
C_EC
0 0
2.5 2.5
5 5
7.5 7.5
10 10
12.5 12.5
15 15
17.5 17.5
Ela
stop
last
Ela
stop
last
Figure 5.11. Interaction between 4-ethylphenol and 4-ethylcatechol for Elastoplast™ descriptor
at 4-ethylguaiacol level 2. C_EC designates the design level of 4-ethylcatechol, and C_EP
indicates the design level of 4-ethylphenol.
34
56
78
9
C_EP
44.555.5
66.577.5
8
C_ISOV
0 0
2.5 2.5
5 5
7.5 7.5
10 10
12.5 12.5
15 15
17.5 17.5
Ela
stop
last
Ela
stop
last
Figure 5.12. Interaction between isovaleric acid and 4-ethylphenol for Elastoplast™ descriptor
for 4-ethylguaiacol level 2. C_ISOV designates the design level of isovaleric acid, and C_EP
indicates the design level of 4-ethylphenol.
As can be seen in these figures, both 4-ethylcatechol and isovaleric acid showed
synergistic effects with 4-ethylphenol in terms of the Elastoplast™ characteristic when 4-
109
ethylguaiacol was at level 2. The Elastoplast™ effect of 4-ethylphenol increased with an
increase in 4-ethylcatechol, as well as with an increase in isovaleric acid. However, this effect is
more severe in the case of 4-ethylcatechol than in the case of isovaleric acid.
4.1.4 55BMedicinal
The results of the ANOVA for the medicinal attribute are shown in XTable 5.19X. There were no
significant interactions for the medicinal descriptor, but three of the compounds, namely 4-
ethylguaiacol (p = 0.0001), 4-ethylcatechol (p = 0.0365) and isovaleric acid (p = 0.0410) could
be identified as statistical main effects. The means of these compounds are shown in XTable
5.20 X and XFigure 5.13 X.
Table 5.19. Results of the ANOVA performed on the sensory data obtained for the medicinal
attribute.
Source DF Mean Square Pr > F
4-EP 4 58.0 0.398
4-EG 3 702 0.0001
4-EC 3 163 0.037
ISOV 3 158 0.041
4-EP*4-EG 1 16.3 0.59
4-EP*4-EC 1 0.15 0.96
4-EG*4-EC 1 5.22 0.76
4-EP*ISOV 1 0.056 0.98
4-EG*ISOV 1 18.0 0.57
4-EC*ISOV 1 23.5 0.52
4-EP*4-EG*4-EC 1 59.9 0.31
4-EP*4-EG*ISOV 1 1.10 0.89
4-EP*4-EC*ISOV 1 22.6 0.53
4-EG*4-EC*ISOV 1 19.7 0.56
4-EP*4-EG*4-EC*ISOV 1 16.8 0.59
Error 495 57.1
As expected, the medicinal attribute increased with an increase in the level of 4-
ethylguaiacol. There was a significant difference between levels 1 and 4, levels 2 and 4 and
between levels 4 and 5. It is interesting to note that the means for 4-ethylguaiacol and medicinal
followed almost exactly the same pattern as the means obtained for 4-ethylguaiacol during the
singular profiling. The range and the means were also similar, as the singular mean for level 1
was 5.5 and 17.2 for level 5. However, the values obtained during combination profiling were
consistently lower than those obtained during singular profiling.
110
Table 5.20. Means for changes in medicinal attribute in Pinotage red wine with changes in
concentration of different compounds.
Level Mean 4-ethylguaiacol Mean 4-ethylcatechol Mean Isovaleric acid
1 4.3 c 11.8 a 9.9 a
2 5.3 c 7.1 b 7.4 a
3 6.9 b c 7.1 b 7.5 a
4 8.8 b 7.0 b 6.7 a
5 14.9 a 5.1 b 3.0 b
Least Significant Difference
(p = 0.05) 3.38 3.38 3.38
1 Values with the same superscript are not significantly different
0
2
4
6
8
10
12
14
16
1 2 3 4 5Level of compound
Inte
nsi
ty o
f M
edic
inal
att
rib
ute
4-EG
4-EC
Isovaleric acid
Figure 5.13. Change in medicinal attribute with different levels of 4-ethylguaiacol, 4-
ethylcatechol and isovaleric acid.
The patterns for 4-ethylcatechol and isovaleric acid were the opposite of that found with
4-ethylguaiacol. Here the means consistently decreased with an increase in the respective
compound. However, level 1 of 4-ethylcatechol was significantly different from all the other
levels, which were not significantly different to one another. Conversely, level 5 of isovaleric acid
was significantly different from the other levels, which were not significantly different from one
another. From this it can be deduced that 4-ethylcatechol suppresses the medicinal descriptor
but only when it is present above detection threshold and that isovaleric acid suppresses the
medicinal descriptor only at level 5. These suppressant effects account for the slightly lower
values found for 4-ethylguaiacol during combination profiling than during singular profiling.
111
4.1.5 56BSmoky/Savoury
The results of the ANOVA for the smoky/savoury characteristic are shown in XTable 5.21X. In
terms of the smoky/savoury characteristic, the only significant effect was 4-ethylguaiacol (p =
0.0002). This is slightly unexpected, as 4-ethylcatechol also has a smoky character associated
with it and therefore some interaction between the two compounds was expected. The overall
effect and means grouping of 4-ethylguaiacol are shown in XTable 5.22X and XFigure 5.14 X.
Table 5.21. ANOVA for sensory data obtained regarding the smoky/savoury attribute.
Source DF Mean Square Pr > F
4-EP 4 20.2 0.73
4-EG 3 265 0.0002
4-EC 3 31.3 0.51
ISOV 3 64.3 0.19
4-EP*4-EG 1 2.00 0.82
4-EP*4-EC 1 2.33 0.81
4-EG*4-EC 1 0.347 0.93
4-EP*ISOV 1 4.50 0.74
4-EG*ISOV 1 14.2 0.55
4-EC*ISOV 1 36.1 0.35
4-EP*4-EG*4-EC 1 0.22 0.94
4-EP*4-EG*ISOV 1 23.3 0.44
4-EP*4-EC*ISOV 1 12.5 0.58
4-EG*4-EC*ISOV 1 80.2 0.16
4-EP*4-EG*4-EC*ISOV 1 11.7 0.59
Error 495 40.6
Table 5.22. Effect of 4-ethylguaiacol on smoky/savoury character of wines.
Level Means
1 6.2 b
2 6.2 b
3 6.8 b
4 8.1 b
5 12.2 a
Least Significant Difference (p = 0.05) 2.85 1 Values with the same superscript are not significantly different
A general trend of an increase of smoky/savoury character was observed with an
increase of 4-ethylguaiacol. However, the only a significant difference was between the first four
levels and level 5. The overall maximum mean was significantly lower than the mean obtained
for smoky character when 4-ethylguaiacol was profiled on its own, and this could probably be
ascribed to a similar reason as the lower levels obtained for the sick-sweet characteristic during
combined profiling.
112
0
2
4
6
8
10
12
14
1 2 3 4 5
Level of 4-ethylguaiacol
Inte
nsi
ty o
f S
avo
ury
att
rib
ute
Figure 5.14. Effect of 4-ethylguaiacol on smoky/savoury character in combination with other
compounds.
4.1.6 57BPungent
The ANOVA for the pungent characteristic is shown in XTable 5.23 X.
Table 5.23. ANOVA on sensory data obtained regarding the pungent attribute.
Source DF Mean Square Pr>F
4-EP 4 266 0.0005
4-EG 3 121 0.078
4-EC 3 47.8 0.43
ISOV 3 609 0.0001
4-EP*4-EG 1 48.3 0.34
4-EP*4-EC 1 11.7 0.64
4-EG*4-EC 1 39.0 0.39
4-EP*ISOV 1 86.7 0.20
4-EG*ISOV 1 396 0.006
4-EC*ISOV 1 45.1 0.36
4-EP*4-EG*4-EC 1 12.5 0.63
4-EP*4-EG*ISOV 1 88.9 0.20
4-EP*4-EC*ISOV 1 6.72 0.72
4-EG*4-EC*ISOV 1 18.0 0.56
4-EP*4-EG*4-EC*ISOV 1 17.0 0.57
Error 495 52.9
It was found that there was a significant interaction between isovaleric acid and 4-
ethylguaiacol (p = 0.0064). This interaction is shown in XFigure 5.15 X (z = 93.03 + 8.79 x + 192.3
113
lny - 2.87 x2 -142.0(lny)2 -6.88xlny + 0.00600x3 +35.67(lny)3 + 58.14 x(lny)2 + 1.43x2lny R2 =
0.73). It can be seen that the pungent characteristic generally increased with an increase in
isovaleric acid concentration. However, there was an interaction where isovaleric acid was at
low levels and 4-ethylguaiacol above its detection threshold. 4-ethylguaiacol enhanced the
pungency effect of below-threshold levels of isovaleric acid when it is around or slightly higher
than detection threshold.
As can also be seen in XTable 5.23X, 4-ethylphenol also displayed a significant main effect
(p = 0.0005). XFigure 5.16 X and XTable 5.24 X show that level 4 (1711 µg/L) differed significantly from
levels 1 (82 µg/L) and 5 (4695 µg/L), but none of the other levels differed significantly. This may
be because 4-ethylphenol has some synergistic effect with isovaleric acid.
01
23
45
67
C_EG
23
45
67
8910
C_ISOV
2.5 2.5
5 5
7.5 7.5
10 10
12.5 12.5
15 15
17.5 17.5
Pun
gen
t
Pun
gen
t
Figure 5.15. Interaction between 4-ethylguaiacol and isovaleric acid for pungent attribute.
C_ISOV designates the design level of isovaleric acid, and C_EG designates the design level of
4-ethylguaiacol.
Table 5.24. The effect of 4-ethylphenol on pungent character of Pinotage red wine.
Level Means
1 4.8 b
2 6.9 a b
3 6.7 a b
4 9.6 a
5 5.6 b
Least Significant Difference (p = 0.05) 3.25 1 Values with the same superscript are not significantly different
114
0
2
4
6
8
10
12
1 2 3 4 5
Level of 4-ethylphenol
Inte
nsi
ty o
f P
un
gen
t d
escr
ipto
r
Figure 5.16. Effect of 4-ethylphenol on the pungent characteristic.
4.1.7 58BOverall effects using different methods of multivariate analysis
The data obtained were investigated using PCA, PARAFAC and PLS. The results obtained from
these different analysis methods are presented in this section. Please note that in this section
reference is made to quadrants. The first quadrant is (+;+), the second (-;+), the third (-;-) and
the fourth (+;-).
61BPrincipal Component Analysis (PCA)
A Principal Component Analysis (PCA) biplot of the overall effects of the different compounds is
shown in XFigure 5.17X. Factor (F) 1 explained approximately 38 % of the total variance, and F2
explained approximately 29.5 % of the total variance, with approximately 67.5% of the total
variance explained by these two components. It can be seen that the descriptors medicinal and
savoury associate with one another (quadrant 2) and sick-sweet and Elastoplast™ associate
with one another (quadrant 1). The pungent descriptor associates strongly with F1, and the
berry-like descriptor dominates quadrant 3. These associations already make sense, as 4-
ethylguaiacol was found to be the strongest causal agent for both the sick-sweet and savoury
characters. The association of sick-sweet with Elastoplast™ may have been due to the
interaction between 4-ethylphenol and 4-ethylcatechol that was present for both these
characteristics. However, this cannot be conclusively stated. The position of the berry-like
descriptor is also as expected, as the presence of each of the individual compounds have been
associated throughout the study with a decrease in berry-like character, and it is therefore
expected that berry-like should not associate with any of the other descriptors. In fact, it is to
115
some extent negatively associated with Elastoplast™ and sick-sweet, and to a lesser extent to
medicinal and savoury.
13332222
2224
22422244
2422
2424
24422444
3133
33133331
3333
33353353
3533
4222
4224
42424244
4422
4424
4442
4444
5333
Berry-like
Sick_sweet
Elastoplast
Pungent
Savoury
Medicinal
-4
-3
-2
-1
0
1
2
3
4
-5 -4 -3 -2 -1 0 1 2 3 4 5 6
F1 (37.92 %)
F2
(29.
53 %
)
Figure 5.17. PCA Biplot showing 25 samples and all descriptors (berry-like, sick-sweet
Elastoplast™, medicinal, savoury, and pungent) used during combination profiling. Groups of
samples scoring highest in specific descriptors are indicated. 67.5% of the variance is explained
by the two components.
XFigure 5.17X also shows several sample groupings. The groups that were identified as
significantly different from the other samples for each descriptor were identified, and circled on
XFigure 5.17X. It is interesting to note that the specific descriptors divide the PCA biplot into
approximately four quadrants. Elastoplast™ spans the first quadrant, medicinal spans the
second quadrant (with the inclusion of three samples that lie on the border of this quadrant),
berry-like spans the third quadrant (with the inclusion of the same three “border” samples) and
the pungent descriptor spans the first and fourth quadrants. The slight overlap of the medicinal
and berry-like characters is an interesting occurrence, as 4-ethylguaiacol was not found to be a
significant contributor to berry-like character, and a negative association would be expected.
However, all three the other compounds caused a decrease in the berry-like descriptor (4-
ethylphenol and 4-ethylcatechol were involved in a significant interaction, and isovaleric acid
was a statistical main effect, see XTable 5.5X, Section X4.1.1 X), explaining the positive association.
The position of pungent in relation to the other descriptors can be explained as follows: it is in
116
the opposite direction to the medicinal and savoury characteristics, but in the same half as the
Elastoplast™, berry-like and sick-sweet characteristics. This may be because 4-ethylguaiacol,
the causal agent for the medicinal characteristic, did not contribute to the pungent characteristic.
The weak association that loading for the pungent attribute shows with that of the Elastoplast™
attribute may be explained by the fact that 4-ethylphenol, a compound associated with the
Elastoplast™ attribute in Chapter 4, was identified as a statistical main effect in terms of the
pungent descriptor.
1333
3133
33133331
33333335
3353
3533 5333
Berry-like
Sick-sweet
Elastoplast
Pungent
Savoury
Medicinal
-5
-4
-3
-2
-1
0
1
2
3
4
5
-5 -4 -3 -2 -1 0 1 2 3 4 5 6
F1 (37.92 %)
F2
(29.
53 %
)
4-EP4-EG4-EC
ISOV
Figure 5.18. PCA biplot of all descriptors (berry-like, sick-sweet, Elastoplast™, medicinal,
savoury, and pungent) and some star and centre samples used during combination profiling
showing change in aroma profile of samples with change from lowest to highest levels of each
compound (4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid). 67.5% of the
variance is explained by the two components.
XFigure 5.18X shows the change in aroma profile with the change from lowest
concentration to highest concentration of the compounds tested. Only the star and centre
samples are annotated. The samples containing the lowest concentrations are shown in italics,
whereas the samples containing the highest concentrations are shown in bold. It can be seen
that for 4-ethylphenol, 4-ethylguaiacol and isovaleric acid, the samples move in the directions of
the relevant descriptors when the concentrations were increased from the centre sample (level
3) to the sample containing level 5 of each compound. In the cases of 4-ethylphenol and 4-
ethylguaiacol, these samples move in the same direction of samples which have their
117
concentrations increased from level 2 to 4 (not shown). It is interesting to note that 4-
ethylcatechol moves in the direction of the berry-like descriptor. However, the reason for this is
not apparent.
A very interesting aspect of XFigure 5.18 X is the angles between the vectors – in other
words the differences in direction of change between samples changing from below what is
considered detection threshold to above detection threshold (from level 1 to level 3) and from
above detection threshold to an extreme level (level 3 to level 5). If no interactions were
present, one would expect movement of the samples to be in the similar direction. However, this
is only the case for isovaleric acid. It is likely that this was due to the base concentration of
isovaleric acid found in all red wines. In the cases of 4-ethylphenol, 4-ethylguaiacol and 4-
ethylcatechol, the movement of the samples from level 1 to level 3 is perpendicular to the
movement from level 3 to level 5. In the cases of 4-ethylphenol and 4-ethylguaiacol – where the
samples move in the direction of the relevant descriptors with an increase from level 3 to level 5
– it implies that an increase in concentration from below detection threshold to above detection
threshold affects the aroma profile, but not in terms of the relevant descriptors.
Another noteworthy aspect of XFigure 5.18 X is the difference in size of these vectors. For
4-ethylphenol, 4-ethylguaiacol and isovaleric acid, the change from level 1 to level 3 is much
smaller than the change from level 3 to level 5. This implies that the effect on the aroma profile
is greater when the concentrations of these compounds are at their extreme levels. The latter is
expected. What is somewhat unexpected, however, is the fact that the change of 4-
ethylcatechol from level 3 to level 5 is smaller than the change from level 1 to level 3. This
means that the presence of 4-ethylcatechol at levels higher than threshold has a large sensory
effect on the overall profile of a wine, but increasing the concentration of 4-ethylcatechol above
this level does not have an additional effect. It is also interesting to note the strong association
of the sample containing the highest level of 4-ethylcatechol (3353) and the berry-like
descriptor.
A final remarkable aspect of XFigure 5.18 X is the similarity of the direction of movement
from levels 1 to 3 for 4-ethylphenol, 4-ethylcatechol and isovaleric acid. 4-ethylguaiacol moves
in a direction that is almost perpendicular to these three compounds and in the opposite
direction to the berry-like descriptor. It is tempting to conclude that 4-ethylguaiacol had the
greatest suppressant effect on berry-like character. Yet, as previously discussed, 4-
ethylguaiacol is neither a main effect for this descriptor, nor part-takes in a significant interaction
for this attribute. The explanation to this observation, however, lies with the medicinal descriptor.
The movement of 4-ethylphenol, 4-ethylcatechol and isovaleric acid from level 1 to level 3 are in
exactly the opposite direction to the medicinal descriptor. The suppressant effects of these
compounds on the medicinal aroma of 4-ethylguaiacol (as described in section X4.1.4 X) therefore
explain these different directions.
118
62BPARAFAC
PARAFAC was performed on the raw dataset that contained the responses of all the judges.
This method was applied as it allows for simple multivariate exploration of the data, and
produces an easily interpretable model from all the sensory responses. The main advantage of
this is that the inherent variability between the judges is taken into account by this model, as it
can be performed on the raw sensory data (Bro et al., 2008).
It was found that three factors produced a model with a core consistency of 95%.
Although a two-factor model gave a higher core consistency (100%), relevant sensory
information could still be extracted from the third factor, and thus the three-factor model was
chosen. This three-factor model explained 46.6% of the variation in the dataset. Note that this
value is lower than the 67% that could be explained by PCA, but due to the difference between
PCA and PARAFAC, it is likely that the excess variation explained by PCA is noise. XFigure 5.19 X
shows the loading plot for factors 1 and 2 of the sensory variables, whereas XFigure 5.21 X shows
the loading plot for factors 2 and 3 for the same set of variables.
From XFigure 5.19 X two variables are identified as having a large effect on the variation of
the dataset, namely Elastoplast™ (factor 1) and berry-like (factor 2). Note that the rest of the
variables are clustered together, indicating that these variables do not have such a large effect
on the overall variation of the dataset.
The importance of Elastoplast™ can be explained by the fact that this descriptor could
easily be identified by all the panel members, and that its character is distinctly different from all
the other descriptors in this study. It is also interesting to note that this characteristic is
commonly linked to Brett character in literature (Chatonnet et al., 1992; Wirz et al., 2004;
Romano et al., 2009). As can be seen in the scores plot ( XFigure 5.20 X), the sample most strongly
associated with the Elastoplast™ descriptor (5333) contained both the highest level of 4-
ethylphenol, and had the highest intensity of the Elastoplast™ attribute. The other sample that
also appears to be “driving” this factor is 4244, which had the second highest mean for the
Elastoplast™ attribute.
The berry-like attribute is one that is suppressed by the presence of all the respective
compounds used during this study. This is not only apparent from this study, but also the one
described in Chapter 4. A decrease in fruitiness is also one of the sensory effects commonly
associated with Brett character (Licker et al., 1999; Fugelsang & Zoecklien, 2003; Ugarte et al.,
2005; Fariña et al., 2007; Cliff & King, 2009). As all the samples contained all four the
compounds, a degree of suppression in berry-like character is expected in all the samples. It is
therefore logical that this attribute drives factor 2, as this descriptor is relevant in all of the
samples used in this study. XFigure 5.20 X shows a group of samples driving the variation in the
direction of the berry-like attribute. These samples are samples 3133, 2244, 2242, and 1333.
This sample grouping is notable for two reasons. Firstly, although this group contains the
119
sample with the highest mean in terms of berry-like character (3133), this group does not
directly correspond to the four samples with the highest level of berry-like character (they are
numbers 1, 4, 6 and 8 in the ranking). The PCA plot, (XFigure 5.17 X) also does not show a similar
grouping to XFigure 5.20X. This indicates that the methodology used for data analysis prior to PCA
(i.e. finding the means of all the judges) confounded important information about this grouping.
XFigure 5.21X shows the sensory loadings plot for factor 2 versus factor 3, and the
corresponding scores plot is shown in XFigure 5.22 X. It can be seen that factor 3 is driven by two
attributes, namely medicinal and pungent. This is logical, as each of these descriptors is linked
to a specific compound used in the study, and drive PC 2 in XFigure 5.17 X. This means that
although these two attributes are not the main causes of variation in the dataset, some of the
overall variation can be attributed to them.
When referring to the scores plot (XFigure 5.22 X), two interesting aspects can be observed.
Firstly, the sample containing the highest level of 4-ethylguaiacol (3533) has a high score in the
latent variable described by the medicinal descriptor. This is logical, as this compound is linked
to the medicinal attribute, and had the highest mean for this attribute. This separation can also
be seen in the PCA plot ( XFigure 5.17 X). There is also a general trend of samples with high levels
of 4-ethylphenol (i.e. sample codes starting with either a 4 or a 5) to fall in quadrant 3. However,
no such pattern could be observed with samples containing high levels of isovaleric acid. This
indicates that these samples tend to associate with the pungent descriptor. This is interesting,
as a general increase in the pungent attribute could be seen with an increase in the level of 4-
ethylphenol (see XFigure 5.16 X). From this, it can be concluded that 4-ethylphenol enhances the
pungency associated with isovaleric acid. This sensory interaction may be the underlying cause
for the conflicting results found in studies regarding isovaleric acid and Brett character
(Fugelsang & Zoecklein, 2003; Romano et al., 2009).
When interpreting XFigure 5.19 X and XFigure 5.21 X, it can be concluded that the sick-sweet
descriptor was not important in the dataset, as none of the three factors identified by the
PARAFAC model showed variation due to this descriptor. As described in Chapter 4, this
attribute mainly comes about from the suppression of the natural berry-like character of the
wine, but may not be as perceived intensely when several other sensory factors are present.
This corresponds to literature, as the sick-sweet attribute has not yet been coupled to Brett
character. In XFigure 5.17 X, the sick-sweet attribute correlates strongly with the Elastoplast™
descriptor, and this figure can easily be interpreted incorrectly by interpreting the trends
exhibited for the Elastoplast™ descriptor as if they applied to the sick-sweet attribute. In this
study, this was prevented by an extensive knowledge of the samples in the study and the
literature involved. However, the results of PARAFAC pose none such dangers of incorrect
interpretation.
120
Figure 5.19. Plot of factor 1 versus factor 2 for sensory loadings in PARAFAC performed on raw
sensory data obtained from profiling samples containing varying levels of Brett-related
compounds. Note that the loadings for the sick-sweet, medicinal and savoury attributes
associate with one another.
Figure 5.20. Scores plot for factor 1 versus factor 2 in the PARAFAC model obtained of raw
data of sensory profiling of samples containing different levels of Brett-related compounds. Note
that the scores are labelled as described in Section X3.2 X.
121
Figure 5.21. Plot of factor 2 versus factor 3 for sensory loadings in PARAFAC performed on raw
sensory data obtained from sensory profiling of all combinations of Brett spoilage compounds
used in this study.
Figure 5.22. Scores plot of factor 2 versus factor 3 in the PARAFAC model obtained from raw
sensory data of samples containing different levels of Brett-related compounds. Note that the
scores are labelled as described in Section X3.2 X.
122
The results found in the PARAFAC appear to give different results than PCA, but the
overall conclusions remain the same. The results from PARAFAC are, however, clearer than
those obtained from PCA, and provide a stronger hierarchy in terms of sensory variables.
PARAFAC was complementary to PCA as it allowed for better interpretation of the overall
dataset. PARAFAC can also aid in preventing incorrect conclusions to be drawn from PCA. This
makes sense as the PARAFAC model is more focussed on modelling the systemic variation and
less likely to model noise (Bro, 1997).
63BPartial Least Squares Regression (PLS)
A Partial Least Squares (PLS-2) analysis was also performed using XLStat (Addinsoft, 2009).
This analysis was not performed to attempt to set up a predictive model, but rather to explore
the relationships between the different levels of (placed in the X-space) and the sensory
descriptors (placed in the Y-space). This approach was taken, since the different levels of the
compounds should cause different sensory qualities.
As the compounds had been shown to have diverse effects at different levels, the
chemical composition was set as a category variable. Five components were modelled.
Component 1 and 2 modelled 42% of the variance, whereas component 2 and 3 modelled 43%
of the variance and component 1 and 3 modelled 37% of the variance. Plots of component 1
versus component 2 and component 1 versus component 3 showed a clear separation of
scores and X-loadings according to whether they were “cube” or “star” samples and could
therefore be related to the design (not shown), and therefore provided no additional information.
It can thus be inferred that component 1 models the design and not the data, and that more
information could be obtained from modelling component 2 versus component 3. This can be
seen in XFigure 5.23 X. Please note that XFigure 5.23 X is flipped along both PC 2 and PC 3 in order
to allow comparison with PCA. The rotation in PCA is of an arbitrary nature.
An interesting aspect of XFigure 5.23X is the fact that the Y-loadings follow a similar pattern
to the loadings when a PCA was performed ( XFigure 5.17 X). Several patterns amongst the X-
loadings were also observed. The X-loadings for the higher levels (levels 4 and 5) of 4-
ethylphenol and 4-ethylguaiacol correlated with the descriptors relevant for those compounds
(Elastoplast™ for 4-ethylphenol, and medicinal and savoury for 4-ethylguaiacol). Level 5 of
isovaleric acid also correlated to the pungent descriptor. These relationships are similar to those
observed in Chapter 4. The X-loadings corresponding to the lowest level (level 1) of all of these
compounds tend to fall in the centre of the biplot, which means that these levels did not have a
large effect on the overall sensory profile. This is not unexpected, as these compounds are all
below detection threshold and are therefore not expected to make much of a difference to the
overall sensory profile. The loadings for 4-ethylcatechol also tend to fall towards the centre of
123
the plot, which means that the 4-ethylcatechol did not have a large overall effect on the sensory
profiles of these wines.
Correlations with t on axes t2 and t3
4-EP-1
4-EP-2
4-EP-3
4-EP-4
4-EP-5
4-EG-3
4-EG-2
4-EG-4
4-EG-1
4-EG-5
4-EC-3
4-EC-2
4-EC-44-EC-1
4-EC-5
ISOV-3
ISOV-2ISOV-4
ISOV-1
ISOV-5
Berry-like
Sick-sweet
Elastoplast
Pungent
Savoury
Medicinal
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
-1-0.75-0.5-0.2500.250.50.751t2
t3
X
Y
Figure 5.23. Plot of X (chemical) and Y (sensory) loadings in components 2 and 3 obtained by
PLS. Note that the scales in this figure appear in the opposite direction. The level of each
compound is designated by the last character in the X-loading label. For example, 4-EP-3
designates level 3 of 4-ethylphenol.
However, some of the X-loadings do not fall in the expected patterns. The first of these
are 4-ethylguaiacol at levels 2 and 3, which correlate negatively with the medicinal descriptor
and appear to form a positive correlation with the pungent descriptor. This may be due to the
enhancement effect the 4-ethylguaiacol was found to have on pungency when present at lower
levels (see section X4.1.6 X). The position of the loading that seems the most arbitrary is that of 4-
ethylphenol level 2, which correlates negatively with the Elastoplast™ descriptor and positively
with the berry-like descriptor. However, this can be explained when the scores of this PLS is
investigated ( XFigure 5.24 X).
As can be seen in XFigure 5.24X, the scores group together according to the design. The
cube samples form four distinct groups that are divided according to their 4-ethylphenol and 4-
ethylguaiacol concentrations. Interestingly, within these groups, a pattern according to the
concentrations of 4-ethylcatechol and isovaleric acid can also be observed. The position of level
2 of 4-ethylphenol on XFigure 5.23X and its association with berry-like character is therefore not
because level 2 of 4-ethylphenol causes berry-like character, but due to the berry-like descriptor
124
associating strongly with the samples containing level 2 of both 4-ethylphenol and 4-
ethylguaiacol.
Correlations on axes t2 and t3
5333
44444442
44244422 4244424242244222
3533
3353
333533333331
3313
3133
24442442
24242422
22442242
22242222
1333
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
-1-0.75-0.5-0.2500.250.50.751t2
t3
XYObs4-EP=24-EP=44-EG=24-EG=4
Berry-like
Savoury
Medicinal
Sick-sweet
Elastoplast
Pungent
Figure 5.24. PLS plot of t2 and t3 containing X-loadings, Y-loadings (sensory descriptors; berry-
like, sick-sweet, Elastoplast™, medicinal, savoury, and pungent) and scores (25 samples).
Grouping according to 4-ethylphenol and 4-ethylguaiacol content is indicated.
4.2 38BConsumer analysis
Of the 100 consumers surveyed, 75 % fell in the age group of between 18 and 25. Sixty-nine
percent of the consumers consumed wine at least once a week, and 91% of the consumers
drank wine at least twice per month. Approximately half of the consumers indicated that they
most often consumed Pinotage. In terms of wine knowledge, 48 consumers considered
themselves novice drinkers, whereas 51 consumers felt they had a fair amount of wine
knowledge. None of the consumers that participated in this study considered themselves wine
experts.
For the total group of consumers, the sample effect was significant (p = 0.0050), which
means that the variation in the dataset could be ascribed to differences between the samples
and not just to the inherent differences between consumers. When the analysis was repeated
on the novice consumers (group A), it was found that sample was not a significant effect.
However, the sample effect was highly significant (p = 0.0051) with the remaining group of
consumers (group B). From this, it can be deduced that the effect of sample in the total group
125
was caused by this group of consumers. The difference in effect can also be ascribed to the fact
that novice wine drinkers respond differently to wine and wine faults than more experienced
wine drinkers (Prescott et al., 2005).
External preference mapping was performed, and can be seen in XFigure 5.25X. This
method first maps the descriptive data in the X-space (using PCA), and then subsequently plots
the preference data onto this map obtained (Tenehaus et al., 2005; Meilgaard et al., 2007). The
samples with the highest and lowest levels of the different compounds (star samples) are shown
in bold. If a sample contained level 4 of a compound, it is presented in colour. Blue indicates 4-
ethylphenol; red indicates 4-ethylguaiacol, yellow 4-ethylcatechol and green isovaleric acid. As
can be seen, the samples that contain level 4 of 4-ethylphenol group together, and the samples
containing level 4 of 4-ethylguaiacol group together. Both these groups fall way from the
direction of liking, and from this it can be interpreted that both these compounds have a
negative effect on liking. It is also interesting to note that liking falls in the opposite direction to
the medicinal descriptor, which is associated with high levels of 4-ethylguaiacol in wine (Chapter
4).
5333
4444
4442
4424
4422
424442424224
4222
3533
33533335
3333
3331 3313
3133
2444
2442
2424
2422
2244
2242
22242222
1333
Liking B
Liking A
Liking A+B
-4
-3
-2
-1
0
1
2
3
4
-4 -3 -2 -1 0 1 2 3 4
F1
F2
Figure 5.25. Preference map 25 of samples obtained during consumer analysis using 100
consumers. A refers to the “inexperienced” consumer group, whereas B refers to the
“experienced” consumer group. Samples containing level 4 of each compound (4-ethylphenol,
4-ethylguaiacol, 4-ethylcatechol and isovaleric acid) are shown. Grouping according to 4-
ethylphenol and 4-ethylguaiacol content is indicated. 67.5% of the variance is explained by the
two components.
126
It was expected that the liking of the consumers would fall in a similar direction than the
berry-like descriptor, as the spoilage compounds all suppress berry-like character to some
degree and wine consumers consider the absence of fault as an important quality characteristic
in wine (Charters & Pettigrew, 2007). This, however, was not the case, as liking fell in a
direction that was not associated with any of the descriptors used in this study (XFigure 5.26 X),
which implies that there is a secondary or tertiary characteristic driving liking in this sample set.
MedicinalSavoury
Pungent
Elastoplast
Sick-sweet
Berry-like
-4
-3
-2
-1
0
1
2
3
4
-4 -3 -2 -1 0 1 2 3 4
F1 (37.92 %)
F2
(29.
53 %
)
Figure 5.26. Scores and loading plot for preference map obtained from the data obtained from
100 consumers showing sensory loadings (berry-like, sick-sweet, Elastoplast™, medicinal,
savoury and pungent). 67.5% of the variance is explained by the two components.
PLS based preference mapping was also performed on the sensory and consumer data
and the map can be seen in XFigure 5.27 X and XFigure 5.28 X. In this method, the sensory
descriptive data is in the X-space, and the consumer data is in the Y-space. The principal
components were therefore determined from the variation present in both these data sets and
not just the variation in the descriptive data, as is the case with external preference mapping.
When comparing the liking (Y) loadings and the descriptive (X) loadings in XFigure 5.27 X,
two interesting aspects were revealed. Firstly, the liking vector for the “experienced” group of
consumers (Group B) lie in exactly the opposite direction of the Elastoplast™ descriptor, which
is commonly associated with Brett character. It can therefore be postulated that consumers with
a higher level of wine knowledge find this characteristic most objectionable. Secondly, the liking
vector for the “inexperienced” group of consumers (Group A) lie in the opposite direction to the
medicinal characteristic, indicating that consumers with a lower degree of wine knowledge find
127
the uncharacteristic medicinal aroma more unpleasant than other wine consumers. These
findings are in line with those of Curtin et al. (2008), who found strong negative correlations
between “medicinal” and “leather” aromas in wine and consumer liking
Correlations with t on axes t1 and t2
Berry-like
Sick-sweet
Elastoplast
PungentSavoury
Medicinal
Liking A+B
Liking A
Liking B
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1
t1
t2
X
Y
Figure 5.27. PLS-based preference map for all samples showing sensory descriptors (berry-
like, sick-sweet, Elastoplast™, medicinal, savoury and pungent) and inexperienced (A) and
experienced (B) consumer groups.
XFigure 5.28 X shows both the scores and the loadings for the PLS done on the consumer
data. The overall liking for the experienced group of consumers fall in an area that contains no
samples, which indicates that none of the samples were liked by the experienced consumers. It
can thus be concluded that Brett tainted wines are more disliked by consumers with a level of
wine expertise than casual wine drinkers.
In XFigure 5.28 X, the scores are divided according to their 4-ethylguaiacol content. It can
be seen that the scores for a level 4-ethylguaiacol level of higher than 4 tend to group together.
Star samples in this group all contained the lowest level of the different compounds. No such
grouping could be observed in terms of any of the other compounds. It could therefore be
inferred that the level of 4-ethylguaiacol is a driver of the liking of the inexperienced consumer
group.
128
Correlations on axes t1 and t2
Medicinal
SavouryPungent
Elastoplast
Sick-sweet
Berry-like
Liking B
Liking A
Liking A+B
5333
3533
3353
333533333331
3313
3133
1333
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1t1
t2
XY4-EG < 34-EG = 34-EG > 3
Figure 5.28. PLS scores and loadings plot for PLS showing all descriptors (berry-like, sick-
sweet, Elastoplast™, medicinal, savoury and pungent), 25 samples and inexperienced (A) and
experienced (B) consumer groups. In this figure, only the loadings for the centre and star
samples are labelled.
It is notable that although such a small amount of variance could be modelled by PLS,
the liking patterns of consumers could be explained a lot better than through external (PCA-
based) preference mapping. This is because PCA in the external preference map is based
solely on the sensory attributes of the samples, and disregards the fact that consumer liking
may not be coupled to a specific sensory attribute, but rather to the manner in which the
sensory attributes relate to one another. Although the number of samples used in this study
made it impossible, is likely that more of the variance in the dataset could be explained if all
consumers were to receive all the samples and could then be mapped individually or via cluster
analysis.
When the liking scores for the different samples of wines are explored in more detail,
some interesting aspects are revealed. The mean liking scores for consumer group B are shown
in XTable 5.25 X.
In XTable 5.25 X it can be seen that the two samples that were least liked are the two
containing the lowest level of 4-ethylphenol (1333) and the highest level of 4-ethylphenol (5333)
respectively. When referring to the positions of these samples on XFigure 5.25 X, it can be seen
that these samples correlate neither to each other nor negatively to Liking. The poor correlation
between the two samples seems obvious as they respectively have the highest and lowest
129
intensity of the Elastoplast™ descriptor, which is strongly associated with PC2 in XFigure 5.17 X.
This is further indication that a property not analysed during descriptive analysis, and therefore
not affecting the positions of samples in XFigure 5.17 X, is the main driver of liking in this sample
set.
Table 5.25. Mean liking scores for consumer group B. Values having the same superscript are
not significantly different from one another.
Sample Liking mean
2422 6.22 a
2244 6.00 a b
2424 5.89 a b c
2222 5.44 a b c d
3331 5.43 a b c d
4422 5.38 a b c d
3335 5.38 a b c d
4442 5.33 a b c d
2442 5.33 a b c d
0000 (control) 5.25 a b c d e
2224 5.25 a b c d e
4444 5.18 a b c d e
3333 5.10 a b c d e f
4244 5.00 a b c d e f
4222 4.88 b c d e f
3133 4.73 b c d e f
2242 4.69 b c d e f
3353 4.67 c d e f
4224 4.57 c d e f
4242 4.54 d e f g
3313 4.36 d e f g
4424 4.30 d e f g
3533 4.00 e f g
2444 3.80 f g
1333 3.80 f g
5333 3. 22 g
Least Significant Difference (p = 0.05)
1.318
With further investigation of XTable 5.25 X, it can be seen that most of the “star” samples fall
towards the bottom of the table, and that the two lowest “cube” samples are 2444 and 4424.
This led to the hypothesis that a concept like balance could be a driver of liking, as this was also
found to be an important quality dimension in wine (Charters & Pettigrew, 2007). However,
balance was not quantified during the descriptive analysis. As a possible indication of “balance”,
the samples were grouped according to the relationship between the levels of the different
compounds. The star samples were considered “unbalanced”, as they contained either a very
high or a very low level of only one compound. Furthermore, samples that contained only one
130
compound at a higher or lower level (such as 2444 or 4424) were considered “unbalanced”.
This divided the sample set into 16 “unbalanced” and 10 “balanced” samples.
A simple analysis was performed on the distribution of these types of samples. The
percentage of the total each of these sample types occurring in the nine most liked and nine
least liked samples was investigated. The choice of nine samples was not arbitrary – the first
nine samples are all significantly different from the last four samples, and the last nine from the
first three samples. Nine samples also comprise approximately 33% of the data set (result not
shown).
It was found that 50% of the “balanced” samples occurred in this top group, and 43% of
the unbalanced samples occurred in the bottom group. This indicates that the concept of
balance may indeed be a driver for consumer acceptance for Brett-tainted samples.
From all these results, it is recommended that future studies regarding preference
mapping of wine taints should not only include quantitative descriptive and hedonic data, but
also a third type of conceptual data. Risvik et al. (2008) have performed studies analysing the
way complex concepts relate to the sensory attributes of several food and fragrance products. A
technique of this type could be adapted for use in wine taints, using known quality parameters
(Charters & Pettigrew, 2007) as well as negative characteristics. This will allow researchers to
obtain information about secondary complex quality characteristics –such as body, complexity
and balance – which are not obtained during descriptive analysis and may be essential in
explaining consumer liking of wines. Inclusion of concepts such as “uncharacteristic of wine”
and “spoiled” will also allow researchers to determine whether a sample is considered “tainted”
– which could not be determined using conventional quantitative descriptive analysis.
5 CONCLUSIONS
Previous results (Chapter 4) showed that 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and
isovaleric acid suppress berry-like character and cause an increase in respectively the
Elastoplast™, medicinal, savoury and pungent descriptors. This is in line with findings in
literature. However, when these compounds are combined in wine, they no longer act as
expected or predicted from other studies done on their detection thresholds. For example,
Chatonnet et al. (1992) found that the detection threshold of 4-ethylphenol in red wine was
lower when combined with 4-ethylguaiacol than when on its own. For this reason, one would
expect an enhancement effect to occur between the sensory effects of these two compounds.
However, it was found that 4-ethylguaiacol suppresses the Elastoplast™ attribute associated
with 4-ethylphenol, and that 4-ethylphenol has a slight suppressant effect on the medicinal
character of 4-ethylguaiacol. Similarly, Romano et al. (2009) found that the presence of
isovaleric acid increased the detection threshold of 4-ethylphenol, and one would therefore
expect the Elastoplast™ descriptor to be suppressed by this compound. Finally, Larcher et al.
131
(2008) postulated that 4-ethylcatechol probably does not have a detrimental effect on the
character of wine. Nonetheless, this study found it to have an enhancement effect on the
Elastoplast™ character associated with 4-ethylphenol. These findings make sense in the light of
the fact that studies done of simple mixtures can generally not be extrapolated to multi-
component mixtures (Brossard et al., 2007).
PARAFAC was used as a complementary tool to PCA, and it was found that the panel
could find the largest differences in the Elastoplast™ and berry-like attributes. This may be
because the Elastoplast™ attribute is severely different from the rest of the attributes profiled,
and the fact that the berry-like attribute was relevant in all the samples profiled. PARAFAC also
revealed that the sick-sweet attribute was of low importance in the samples. Additionally,
PARAFAC identified a group of berry-like samples that could not be identified by either
univariate statistics or PCA, and could aid in the explanation of the nature of the effect of 4-
ethylphenol on the pungent attribute. It is therefore recommended that PARAFAC be used to
complement PCA in future sensory studies.
An attempt was made to map the preference of consumers for wines containing
compounds related to Brett character, and three important findings were made. Firstly, some of
the tainted samples were found to be more acceptable to consumers than the base wine (XTable
5.25 X). This was expected, as it is generally accepted that Brett character adds complexity to
wine (Saurez et al., 2007), which can increase the quality of wine (Charters & Pettigrew, 2007).
Secondly, inexperienced wine consumers did not distinguish between the samples in terms of
liking, whereas more experienced consumers detected such differences. This is in line with
findings of other studies dealing with wine taints (Prescott et al., 2005), and it is recommended
that further studies regarding consumers and wine faults such as Brett character should take
this into consideration. Thirdly, it was found that the sample containing the highest level of 4-
ethylphenol was liked the least by consumers, and it can therefore be said that samples
containing a high level of Brett character are objectionable. This is in line with the findings of
Lattey et al. (2007) and Curtin et al. (2008), who also found samples with the highest degree of
Brett character least liked by consumers.
However, the consumer analysis failed to accurately map the consumer liking of these
tainted samples according to sensory descriptors or concentrations of spoilage compounds. It is
speculated that consumer liking of these samples is driven by a secondary sensory
characteristic such as wine balance or complexity. It is therefore recommended that future
studies of this type include conceptual analysis on wines.
This study further underlines the fact that sensory Brett character is not as simple as it
seems, and that the sensory effect is not caused by one or two chemical compounds acting by
themselves, but is rather the result of the interaction of these compounds. An interesting finding
is the fact that the Elastoplast™ descriptor is not only affected by 4-ethylphenol or 4-ethylphenol
and 4-ethylguaiacol, but also by 4-ethylcatechol and isovaleric acid. It is therefore
132
recommended that future studies on Brett character should focus on all four these compounds,
and not only on 4-ethylphenol and 4-ethylguaiacol. This is of particular importance in the South
African wine industry, as Pinotage contains excessive levels of the precursors of 4-ethylcatechol
(de Villiers et al., 2005), and is therefore more susceptible to elevated levels of this compound.
Although it is tempting to extrapolate these exciting findings into the real world situation,
it still only scratches the surface of the situation that is sensory Brett character. Sensory effects
in wine are affected by a variety of factors such as cultivar, wine style, alcohol content and the
degree of inherent fruitiness the wine possesses (Le Berre et al., 2007; Escudero et al., 2007).
This has also been found to be the case with Brett character (Norris, 2004; Suarez et al., 2007).
As this study highlighted the interactions between these compounds, the next step is to further
investigate these interactions in a variety on wines, as well as in a variety of different
combinations. Should this be done, Brett character can be modelled more successfully, and
chemical analysis could become a better tool for the prediction of sensory Brett character.
6 REFERENCES
Atanasova, B., Thomas-Danguim, T., Langious, D., Nickluas, S., Chabanet, C., & Etievant, P.
(2005a) Perception of wine fruity and woody notes: influence of peri-threshold odorants.
Food Quality and Preference. 16, 504 - 510.
Atanasova, B., Thomas-Danguin, T., Chabanet, C., Langious, D., Nickluas, S., & Etievant, P.
(2005b) Perceptual interactions in odour mixtures: odour quality in binary mixtures of
woody and fruity wine odorants. Chemical Senses. 30, 209 - 217.
Atanasova, B., Thomas-Danguin, T., Langious, D., Nickluas, S., & Etievant, P. (2004)
Perceptual interactions between fruity and woody notes of wine. Flavour and Fragrance
Journal. 19, 476 - 482.
Bro, R. (1997) PARAFAC. Tutorial and applications. Chemometrics and Intelligent Laboratory
Systems. 38, 149 - 171.
Bro, R., Qannari, E. M., Kiers, H. A. L., Næs, T. & Frøst, M. B. (2008) Multi-way models for
sensory profiling data. Journal of Chemometrics. 22, 36 - 45.
Brodin, M., Moeller, P. & Olsson, M. (2007) Are we mixing odorants or odors? (Abstract)
Chemical Senses. 32, A19.
Brossard, C., Rousseau, F. & Dumont, J. (2007) Perceptual interactions between characteristic
notes smelled above aqueous solutions of odorant mixtures. Chemical Senses. 32, 319 -
327.
Charters, S. & Pettigrew, S. (2007) The dimensions of wine quality. Food Quality and
Preference. 15, 997 - 1009.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
133
Cheng, H. & Friis, A. (2008) Expansion of the whole wheat flour extrusion. Paper presented at
5th International Conference on Simulation and Modelling in the Food and Bio-Industry -
FoodSim 2008, Dublin, Ireland, June 26-28, 2008. .
Cliff, M. A. & King, M. C. (2009) Influence of serving temperature and wine type on perception of
ethyl acetate and 4-ethylphenol in wine. Journal of Wine Research. 20, 45 - 52.
Cocchi, M., Bro, R., Duranye, C., Manzini, D., Marchetti, A., Saccani, F., Singhinolfi, S. & Ulrici,
A. (2006) Analysis of sensory data of Aceto Balsamica Tradizionale di Modena (ABTM)
of different ageing by application of PARAFAC models. Food Quality and Preference.
17, 419 - 428.
Curtin, C. Bramley, B. Cowey, G. Holdstock, M. Kennedy, E. Lattey, K. Coulter, A. Henschke, P.
Francis, L. Godden, P. Sensory perceptions of 'Brett' and relationship to consumer
preference. Blair, R.J.; Williams, P.J.; Pretorius, I.S. Proceedings of the thirteenth
Australian wine industry technical conference, 29 July-2 August 2007, Adelaide, SA. :
207-211; 2008.
De Villiers, A., Majek, P., Lynen, F., Crouch, A., Lauer, H. & Sandra, P. (2005) Classification of
South African red and white wines according to grape variety based on the non-coloured
phenolic content. European Food Research and Technology. 221, 520 - 528.
Esbensen, K. H. (2002). Introduction to experimental design. In: Multivariate Data Analysis, 5th
ed. Pp 361 - 446. Osla, Norway: CAMO Process AS
Escudero, A., Campo, E., Fariña, L., Cacho, J. & Ferreira, V. (2007) Analytical characterization
of the aroma of five premium red wines. Insights into the role of odor families as the
concept of Fruitiness of Wines. Journal of Agricultural and Food Chemistry. 55, 4501 -
4510.
Eteiévant, P. X. (1981) Volatile phenol determination in wine. Journal of Agricultural and Food
Chemistry. 29, 65 - 67.
Eteiévant, P. X., Issanchou, S. N., Marie, S., Ducruet, V. & Flanzy, C. (1989) Sensory impact of
volatile phenols on red wine aroma: influence of carbonic maceration time and storage.
Sciences des aliments. 9, 19 - 31.
Fariña, L., Boido, E., Carrau, F. & Dellacassa, E. (2007) Determination of volatile phenols in red
wines by dispersive liquid-liquid microextraction and gas chromatography-mass
spectrometry detection. Journal of Chromatography A. 1157, 46 - 50.
Francis, I. L. & Newton, J. L (2005) Determining wine aroma from compositional data. In: AWRI
– Advances in wine science (edited by R. J. Blair, M. E. Francis & I. S. Pretorius A. L.),
Pp 201 - 212. Glen Osmond, Australia: Australian Wine Research Institute.
Fugelsang, K. C. & Zoeklein, B. W. (2003) Population dynamics and effects of Brettanomyces
bruxellensis strains on Pinot noir (Vitis vinifera L.) wines. American Journal of Enology
and Viticulture. 54, 294 - 300.
134
Glass, G.V., Peckham, P.D. & Sanders, J.R. (1972). Consequences of failure to meet
assumptions underlying the fixed effects analyses of variance and covariance. Review of
Educational Research, 42, 237-288.
Grosh, W. (2001) Evaluation of the key odorants of foods by dilution experiments, aroma
models and omission. Chemical Senses. 26, 533 - 545.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11-13.
Hesford, F., Schneider, K., Porret, N. & Gafner, J. (2004) Identification and analysis of 4-
ethylcatechol in wines tainted by Brettanomyces off-flavor (Abstract). American Journal
of Enology and Viticulture. 55, 304A.
Hughson, A. L. & Boakes, R. A. (2002) The knowing nose: the role of knowledge in wine
expertise. Food Quality and Preference. 13, 463 - 472.
Jinks, A. & Laing, D. G. (1999) Temporal processing reveals a mechanism for limiting the
capacity of humans to analyse odor mixtures. Cognitive Brain Research. 8, 311 - 324.
Laing, D. G. (1988) Relationship between the differential adsorption of odorants by the olfactory
mucus and their perception in mixtures. Chemical Senses. 13, 463 - 471.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Laska, M. & Hudson, R. (1991) A comparison of the detection thresholds of odour mixtures and
their components. Chemical Senses. 16, 651 - 662.
Lattey, K. A., Bramley, B. R., Francis, I. L., Herderich, M. J. & Pretorius, S. (2007) Wine quality
and consumer preferences: understanding consumer needs. Australian and New
Zealand Wine Industry Journal. 22, 31 - 39.
Lawless, H. T. & Heymann, H. (1998). Sensory evaluation of food, principles and practice.
New York, USA: Chapman and Hall.
Lawless, H. T. (1999) Descriptive analysis of complex odors: reality, model or illusion. Food
Quality and Preference. 10, 325 - 332.
Le Berre, E., Atanasova, B., Langloi, Etiéviant, P. & Thamas-Danguin, T. (2007) Impact of
ethanol on the perception of wine odorant mixtures. Food Quality and Preference. 18,
901 - 908.
Le Berre, E., Béno, N., Ishii, A., Chabanet, C., Etiéviant, P. & Thomas-Danguin, T. (2008b) Just
noticable differences in component concentrations modify the odor quality of a blending
mixture. Chemical Senses. 33, 489 - 395.
Le Berre, E., Thomas-Danguin, T., Béno, N., Coureaud, G., Etiéviant, P. & Prescott, J. (2008a)
Perceptual processing strategy and exposure influence the perception of odor mixtures.
Chemical Senses. 33, 193 - 199.
135
Licker, J. L, Acree, T. E. & Henick-Kling, T. (1999) What is "Brett" (Brettanomyces) flavour? A
preliminary investigation. In: Chemistry of Wine Flavour. ACS Symposium Series (edited
by A. L. Waterhouse & S. E. Ebeler), Pp 96 - 115. Washington DC: American Chemical
Society.
Livermore, A. & Laing, D. G. (1996) Influence of training and experience on the perception of
multicomponent odor mixtures. Journal of Experimental Psychology. 22, 267 - 277.
Masino, F., Foca, G., Ulrici, A., Arru, L. & Antonelli, A. (2008) A chemometric study of pesto
sauce appearance and of its relation to pigment concentration. Journal of the Science of
Food and Agriculture. 88, 1335 - 1343.
Meilgaard, M. C., Civille, G. V. & Carr, B. T. (2007) Sensory Evaluation Techniques. Baco
Raton, USA: CRC Press.
Mirhosseini, H & Tan, C. P. (2009) Response surface methodology and multivariate analysis of
equilibrium headspace concentration of orange beverage emulsion as function of
emulsion composition and structure. Food Chemistry. 115, 324-333.
Mirhosseini, H., Tan, C. P., Hamid, N. S. A. & Yusof, S. (2008) Effect of Arabic gum, xanthum
gum and orange oil on flavor release from diluted orange beverage emulsion. Food
Chemistry. 107, 1161-1172.
Noble, A.C., Arnold, R.A., Buechsenstein, J., Leach, E. J., Schmidt, J. O. & Stern, P. M. (1987)
Modification of a standardized system of wine aroma terminology. American Journal of
Enology and Viticulture. 38, 143 - 146.
Norris, L. (2004) Unraveling the mystery of Brettanomyces flavour (Abstract). American Journal
of Enology and Viticulture. 55, 304A.
Oelofse, A., Lonvaud-Funel, A. & du Toit, M. (2009) Molecular identification of Brettanomyces
bruxellensis strains isolated from red wines and volatile phenol production. Food
Microbiology. 26, 377 - 385.
Pravoda, V., Boucon, C., de Jong, S., Walczak, B. & Massart, D. L. (2002) Three-way principal
component analysis applied to food analysis: an example. Analytica Chimica Acta. 462,
133 - 148.
Prescott, J., Norris, L., Kunst, M. & Kim, S. (2005) Estimating a "consumer rejection threshold"
for cork taint in white wine. Food Quality and Preference. 16, 345 - 349.
Risvik, E. (2008) From Simplicity to Complexity in Sensory Science. Paper presented SAAFost
Sensory Symposium, Cape Town, South Africa, August 13, 2008.
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of “Brett character”. Food Chemistry. 114, 15 - 19.
Saurez, R., Saurez-Lepe, J. A., Morata, A. & Calderon, F. (2007) The production of
ethylphenols in wine by yeasts of the genera Brettanomyces and Dekkera: A review.
Food Chemistry. 102, 10-21.
136
Shapiro, S. S. & Wilk, M. B. (1965). An analysis of variance test for normality (complete
samples). Biometrika, 52, 591-611.
Tenenhaus, M., Pagès J., Ambroisine, L. & Guinot, C. (2004) PLS methodology to study
relationships between hedonic judgements and product characteristics. Food Quality
and Preference. 16, 315 - 325.
Ugarte, P., Agosin, E., Bordeu, E. & Villalobos, J. I. (2005) Reduction of 4-ethylphenol and 4-
ethylguaiacol concentration in red wines using reverse osmosis and adsorption.
American Journal of Enology and Viticulture. 56, 30 – 36.
Venkateshwarlu, G., Let, M. B., Meyer, A. S. & Jacobs, C. (2004) Modeling the sensory impact
of defined combinations of volatile lipid oxidation products on fishy and metallic off-
flavours. Journal of Agricultural and Food Chemistry. 52, 1635 - 1641.
Wirz, D. O., Heymann, H. & Bisson, L. F. (2004) Descriptive analysis of Brettanomyces-infected
Cabernet Sauvignon wines. American Journal of Enology and Viticulture. 55, 303A.
Wise, B. M., Gallagher, N. B., Bro, R., Shaver, J. M., Windig, W., and Koch, R. S. PLS Toolbox.
[Version 4.3]. 2009. Eigenvector Research Inc., Manson, WA.
137
Chapter 6: 7BExplorative investigation into the incidence of eight Brettanomyces-related spoilage compounds in a selection of South
African red wines
1 INTRODUCTION........................................................................................................... 138
2 MATERIALS AND METHODS...................................................................................... 138
2.1 Samples..................................................................................................................... 138
2.2 Chemical analyses ................................................................................................... 138
2.3 Statistical analysis ................................................................................................... 140
3 RESULTS AND DISCUSSION ..................................................................................... 140
3.1 Quantitative results .................................................................................................. 140
3.2 Relationships between compound levels .............................................................. 143
4 CONCLUSIONS............................................................................................................ 146
5 REFERENCES.............................................................................................................. 146
138
1 INTRODUCTION
The chemical analysis of 4-ethylphenol and 4-ethylguaiacol is commonly used for the diagnosis
of wines spoiled by Brettanomyces. However, poor correlations have been found between the
levels of these compounds and sensory attributes associated with this yeast (Fugelsang &
Zoecklein, 2003; Romano et al, 2009). Recent studies have tended to not only include 4-
ethylphenol and 4-ethylguaiacol, but several other compounds as well. These include
predominantly 4-ethylcatechol (Curtin et al., 2008), as well as isovaleric acid (Romano et al.,
2009).
As 4-ethylcatechol was only linked to the spoilage of red wine in 2004, very few studies
have been performed investigating this compound. This is partly due to the fact that the
chemical characteristics of 4-ethylcatechol necessitates the use of a different method of
chemical analysis than the “classical” methods used for the other volatile phenols. 4-
ethylcatechol may be analysed by gas chromatography (GC) preceded by a derivitisation step
(Hesford & Schneider, 2004; Carillo & Tena, 2007) or by means of high performance liquid
chromatography (HPLC) (Larcher et al., 2008)
The aim of this study was to determine the concentrations of eight Brett-related spoilage
compounds (4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol, 4-vinylphenol, 4-vinylguaiacol,
isovaleric acid, isobutyric acid and acetic acid) in a selection of South African red wines. The
sample set was on purpose selected to contain a large proportion of wines potentially spoiled by
Brettanomyces, especially with the aim to investigate the occurrence of 4-ethylcatechol.
2 MATERIALS AND METHODS
2.1 39BSamples
Thirty-four red wine samples were sourced from the South African wine industry. Of these thirty-
two samples, twelve were obtained from a wine analysis facility, and were considered to be
spoiled with Brettanomyces. Twenty wines were directly obtained from South African wine
cellars, of which ten were already on the market. The remaining two South African wines were
the wines used for the sensory analysis component of this study.
2.2 40BChemical analyses
Analysis of 4-ethylphenol, 4-ethylguaiacol, 4-vinylphenol, 4-vinylguaiacol, isovaleric acid,
isobutyric acid and acetic acid was performed using GC-MS. Extraction was performed using 2
139
mL diethyl ether in 10 mL of wine, using 2,3- dimethyl phenol (50 µg/L) (Sigma Aldrich,
Germany) as internal standard The samples were subsequently sonicated for 30 minutes
(shaken at 5 minute intervals) and then dried over sodium sulphate. The organic phase of each
extraction was collected, and 2 µL of the extract was injected into an Agilent 5890 GC-MS. The
column employed was a DB-FFAP (60m x 320.0 µm x 0.5 µm column) and the carrier gas was
helium. The injector (split/splitless) was heated to 260 ºC with a splitless time of 1 minute and a
split flow of 1.2 mL/min. The oven temperature was increased after injection from 40 ºC at a rate
of 20 ºC/min up to 150 ºC and then at 5 ºC/min up to 240 ºC, where it was held for 8 minutes.
MS quantisation was performed in selected ion monitoring (SIM) mode. The relevant ions are
shown in XTable 6.1X.
Table 6.1. Ions used for MS analysis in SIM mode.
Compound Ions
4-ethylphenol 107, 122
4-ethylguaiacol 137, 152
4-vinylphenol 91, 120
4-vinylguaiacol 135, 150
Isovaleric acid 60, 69, 87
Isobutyric acid 73, 88, 55
Acetic acid 46, 60, 61
2,3-dimethylphenol 107, 122
The analysis for 4-ethylcatechol was performed using HPLC with tandem mass
spectrometric (MS/MS) detection. This method was chosen as difficulties were experienced in
developing a GC method suitable for analysis of this compound. A Waters Alliance 2695 liquid
chromatograph (Waters Corporation, Milford, U. S. A.) equipped with a Waters API Quattro
Micro tandem quadruple mass spectrometric detector was used for this analysis. Sample
extracts were separated by reversed phase liquid chromatography utilizing an acetonitrile and
water gradient and a C18 column (Waters Xbrige, 2.1 x 50 mm with guard) The gradient started
at 5% acetonitrile, increased to 90% in 6 minutes, followed by a clean-up step consisting of 95%
water for 4 minutes. The flow was maintained at 0.4 mL/min throughout. Negative electrospray
ionization was performed using the following optimized parameters: cone voltage 15 V, capillary
voltage, 3.5 kV, source and desolvation temperatures 100 ºC and 400 ºC respectively. Nitrogen
was used as desolvation and cone gas at flow rates of 400 L/h and 50 L/h respectively. The
mass spectrometer was operated in multiple reaction monitoring mode. Acquisition parameters
are given in XTable 6.2X.
The calibration for this method was performed using standards done between 10 µg/L
and 500 µg/L, as well as wine samples up to 500 µg/L. The R2 for the calibration curve was
0.99.
140
Table 6.2. Parameters for aquisition of MRM data.
Parent ion (Da) Collision Energy (eV) Daughter ion (Da)
119.0 20 93.0
122.0 15 108.0
136.8 15 122.0
149.0 15 134.20
151.0 15 136.20
2.3 41BStatistical analysis
Principal component analysis (PCA) was performed on quantitative data for the analysed South
African wines. Linear regression was performed to investigate the relationships between the
compounds between different sets of compounds between the data. The mean, median,
minimum and maximum values were also calculated for each compound. All these analyses
were performed using the XLStat software (Version 2009.5.0.1 Addinsoft, SARL, Paris, France).
3 RESULTS AND DISCUSSION
3.1 42BQuantitative results
A summary of the overall results can be seen in XTable 6.3 X, which reveals some interesting
aspects of the dataset. Firstly, the maximum level of 4-ethylphenol found in the dataset was
16330 µg/L. This value is extremely high, as it exceeds the value mentioned in literature
(Francis & Newton, 2005) by approximately four times. It is also four times as high as the value
used in this study (Chapters 4 and 5). However, this wine was identified as being excessively
spoiled. Another interesting aspect is the fact the maximum level of 4-ethylcatechol found was
only 227.7 µg/L, in spite of the fact that some of the wines analysed were severely spoiled. This
value is lower than the detection threshold found in this study (385 µg/L, see Chapter 3), as well
as the maximum value of 1610 µg/L found by Larcher et al. (2008). However, the median, first
quartile (25th percentile) and third quartile (75th percentile) showed good agreement with the
latter study.
XFigure 6.1X and XFigure 6.2 X show the results of PCA performed on the entire dataset. In
these figures, all the compounds except for 4-ethylcatechol are highly correlated, and the
variation in their levels explain most of the variation in the dataset (PC1 76%). It is interesting to
note that loadings for 4-ethylphenol and 4-ethylguaiacol lie close together, as do the loadings
for isobutyric and isovaleric acid. However, the loading value for 4-ethylcatechol does not
141
correlate with the other variables. From this it can be inferred that the levels of 4-ethylcatechol in
this sample set are not correlated to the levels of 4-ethylphenol and 4-ethylguaiacol for the
analysed samples. Furthermore, the six samples to the right of XFigure 6.2 X contain the highest
levels of Brett spoilage compounds, and it can therefore be said that PC 1 explains Brett
spoilage. 4-ethylcatechol explains most of the variation in PC 2. It can thus be concluded that
the levels of 4-ethylcatechol in this sample set cannot be directly linked to Brett spoilage.
Table 6.3. Summary of results of chemical analyses of Brettanomyces derived compounds of
34 South African red wines.
Mean
25th percentile
Median 75th
percentile95th
percentile Maximum
4-ethylphenol (µg/L) 1591 54 404 687 5368 16330
4-ethylguaiacol (µg/L) 146 11 39 71 419 1236
4-ethylcatechol (µg/L) 47 8 30 56 114 228
4-vinylphenol (µg/L) 6278 785 1379 3310 19063 58685
4-vinylguaiacol (µg/L) 283 45 70 178 1301 1871
Isovaleric acid (µg/L) 2780 485 838 2010 11349 18995
Isobutyric acid (µg/L) 759 133 217 478 3196 5103
Acetic acid (g/L) 2.80 0.44 0.83 1.59 11.82 14.98
Variables (axes F1 and F2: 89.36 %)
4VP
Acetic acid
Isobutyric acidIsovaleric acid
4-EG
4-EP
4VG
4-EC
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1
F1 (75.33 %)
F2
(14.
03 %
)
Figure 6.1. Principal component analysis loadings plot on for the chemical data obtained from
34 red wine samples. Compounds included are 4-ethylphenol (4-EP), 4-ethylguaiacol (4-EG) 4-
ethylcatechol (4-EC), 4-vinylphenol (4-VP), 4-vinylguaiacol (4-VG), isovaleric acid, isobutyric
acid and acetic acid.
142
A second principal component analysis was performed on a selected sample set ( XFigure
6.3 X). The samples in this set were selected by determining the total value for 4-ethylphenol and
4-ethylguaiacol in the sample. If this number fell between 426 µg/L (the diagnostic value
determined by Chatonnet et al., 1992) and 5000 µg/L (the maximum combined concentration
used in this study), the sample was included in the dataset. This was done to only include
samples that were spoiled by Brettanomyces, but to exclude those that were severely spoiled.
This allowed for a more detailed investigation into the smaller differences in chemical
composition between the spoiled samples. A PCA biplot for this analysis is shown in XFigure 6.3X.
Biplot (axes F1 and F2: 89.36 %)
C7
B5
B4
S8
C6
M4
S7
M3
P8B3
S6
S5
S4
S3
S2
C5
C4M2
P7
P6P5
P4P3
P2
B2
C3
P1C2C1S1 M1
B1
4-EC
4VG
4-EP4-EG
Isovaleric acidIsobutyric acid
Acetic acid
4VP
-5
0
5
-5 0 5 10 15 20
F1 (75.33 %)
F2
(14.
03 %
)
Figure 6.2. Principal component analysis biplot on chemical data obtained from 34 red wine
samples. Sample names refer to the cultivar contained in the sample (B – Blend; C – Cabernet
sauvignon; M – Merlot; P – Pinotage; S – Shiraz).
The results obtained do suggest that cultivar has an influence in the amount of 4-
ethylcatechol present in a sample, as the three Pinotage samples in the biplot tend to associate
with the loading for 4-ethylcatechol. This suggests that the hypothesis that this cultivar is more
subject to higher levels of 4-ethylcatechol due to its higher levels of precursors may be valid.
143
Biplot (axes F1 and F2: 85.20 %)
B6
B4
C6
M3
B3
S6S4
C4
M2
P6
P5
P2
B2
4-EC
4VG
4-EP4-EG
Isovaleric acid
Isobutyric acid
Acetic acid
4VP
-5
0
5
-7 -2 3 8 13
F1 (62.71 %)
F2
(22.
50 %
)
Figure 6.3. Principal component analysis of samples where the sum of 4-ethylphenol and 4-
ethylguaiacol was between 426 µg/L and 5000 µg/L. Sample names refer to the cultivar
contained in the sample (B – Blend; C – Cabernet sauvignon; M – Merlot; P – Pinotage; S –
Shiraz). Pinotage samples, as well as Pinotage-based blends are highlighted in red.
3.2 43BRelationships between compound levels
Linear regression analysis was performed on sets of compounds shown in XTable 6.3 X. The
results are summarised in XTable 6.4X.
Table 6.4. Results of regression analysis of the quantitative data for all compounds.
Compounds P for regression R2 value
4-EP and 4-EG < 0.0001 0.966
4-EP and 4-EC 0.785 0.003
4-EP and isovaleric acid < 0.0001 0.888
4-EP and isobutyric acid < 0.0001 0.900
4-EG and 4-EC 0.816 0.002
4-EG and isovaleric acid < 0.0001 0.854
4-EG and isobutyric acid < 0.0001 0.874
Isovaleric acid and isobutyric acid < 0.0001 0.996
The first interesting aspect of XTable 6.4X is the fact that significant correlations were
found between all the sets of variables except for the ones containing 4-ethylcatechol. This is
144
particularly unexpected since 4-ethylcatechol is produced by the same enzymatic pathway as 4-
ethylphenol and 4-ethylguaiacol, and a relationship is therefore expected. However, Larcher et
al. (2008) pointed out that cultivar has a significant effect on the production of 4-ethylcatechol,
which might explain the results in XTable 6.4X.
XFigure 6.4 X, XFigure 6.5 X and XFigure 6.6 X show the three most significant relationships listed
in XTable 6.4X. As it can be seen, there is a strong linear relationship (R2 = 0.97, XFigure 6.4 X)
between the level of 4-ethylphenol and 4-ethylguaiacol in the sample set. This is expected, as
these two compounds are produced by the same enzymatic pathway. A similarly strong
relationship (R2 = 0.97 XFigure 6.6 X) is exhibited between the levels of isovaleric acid and
isobutyric acid. This is also likely that they are formed by the same biochemical pathway
(Harwood & Canale-Parola, 1981). However, the most interesting aspect is depicted in XFigure
6.5 X. This figure shows a strong correlation (R2 = 0.88) between the level of 4-ethylphenol and
isovaleric acid in the samples. This relationship could also be observed between other 4-
ethylphenol and isobutyric acid, as well as 4-ethylguaiacol and both isovaleric and isobutyric
acid (XTable 6.4 X). This relationship is in agreement with the findings of Romano et al. (2009), and
is further evidence that all four these compounds contribute to the sensory effect which is Brett
character.
Regression of 4-EG by 4-EP (R²=0.966)
-200
0
200
400
600
800
1000
1200
1400
1600
0 5000 10000 15000 20000
4EP
4EG
Active Model
Conf. interval (Mean 95%) Conf. interval (Obs. 95%)
Figure 6.4. Relationship between 4-ethylphenol and 4-ethylguaiacol content of selected South
African red wines. The curve has the following equation: 4-EG = 10.50+0.085*4-EP
145
Regression of Isovaleric acid by 4-EP (R²=0.888)
-5000
0
5000
10000
15000
20000
25000
0 5000 10000 15000 20000
4-EP
Iso
vale
rica
cid
Active Model
Conf. interval (Mean 95%) Conf. interval (Obs. 95%)
Figure 6.5. Relationship between 4-ethylphenol and isovaleric acid in selected South African
red wines. The curve satisfies the following equation: Isovaleric acid = 618+1.14*4EP
Regression of Isobutyric acid by Isovaleric acid (R²=0.996)
-1000
0
1000
2000
3000
4000
5000
6000
0 5000 10000 15000 20000
Isovaleric acid
Iso
bu
tyri
c ac
id
Active Model
Conf. interval (Mean 95%) Conf. interval (Obs. 95%)
Figure 6.6. Relationship between isovaleric acid and isobutyric acid in selected South African
red wines. The curve has the following equation: Isobutyric acid = -12.34 +0.26 * Isovaleric acid
There are two possible explanations for the poor correlations found between the levels
of 4-ethylcatechol and the other two ethylphenols used in this study. The first, which has already
been mentioned, is the effect of cultivar. It is likely that differences in the levels of precursors
between the respective cultivars lead to different levels of the spoilage compounds. This would
be in agreement with the findings of Hesford et al. (2004) and Larcher et al. (2008), who
investigated these effects. However, in order to prove this hypothesis, a more comprehensive
study containing more representative numbers of each cultivar needs to be undertaken. The
146
second possible cause of this phenomenon is microbiological. In their review on Brettanomyces
in wine, Renouf et al. (2007) presents the theory that this yeast decarboxilates the
hydrocinnamic acids in order to protect itself against their toxic effects. Should caffeic acid show
a different (lower) toxicity towards Brettanomyces than p-coumaric and ferulic acid, it may result
in a lower production of 4-ethylcatechol.
4 CONCLUSIONS
This study was the first to investigate the levels of 4-ethylcatechol found in selected South
African red wines. It was found that the levels of 4-ethylcatechol are not related to the levels of
4-ethylguaiacol and 4-ethylphenol found, which is likely that this is due to the cultivar effect.
However, strong correlations were found between the levels of 4-ethylphenol and both
isovaleric and isobutyric acids. It can therefore be concluded that these carboxylic acids are
formed by Brettanomyces alongside the ethylphenols. In order to investigate the effect of
cultivar on the levels of 4-ethylcatechol produced by Brettanomyces, it is recommended that a
larger study should be undertaken investigating a more representative sample set.
5 REFERENCES
Carrillo, J. D. & Tena, M. T. (2007) Determination of ethylphenols in wine by in situ derivitisation
and headspace solid-phase microextraction-gas chromatography-mass spectrometry.
Analytical and Bioanalytical Chemistry. 387, 2547 - 2558.
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
Curtin, C. Bramley, B. Cowey, G. Holdstock, M. Kennedy, E. Lattey, K. Coulter, A. Henschke, P.
Francis, L. Godden, P. Sensory perceptions of 'Brett' and relationship to consumer
preference. Blair, R.J.; Williams, P.J.; Pretorius, I.S. Proceedings of the thirteenth
Australian wine industry technical conference, 29 July-2 August 2007, Adelaide, SA. :
207-211; 2008.
Francis, I. L. & Newton, J. L (2005) Determining wine aroma from compositional data. In: AWRI
– Advances in wine science (edited by R. J. Blair, M. E. Francis & I. S. Pretorius A. L.),
Pp 201 - 212. Glen Osmond, Australia: Australian Wine Research Institute.
Fugelsang, K. C. & Zoeklein, B. W. (2003) Population dynamics and effects of Brettanomyces
bruxellensis strains on Pinot noir (Vitis vinifera L.) wines. American Journal of Enology
and Viticulture. 54, 294 - 300.
147
Harwood C. S. & Canale-Parola, E. (1981) Adenosine-5'-Triphosphate-yielding pathways of
branched-chain amino acid fermentation by a marine Spirochete. Journal of
Bacteriology. 148, 117 - 123.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11-13.
Hesford, F., Schneider, K., Porret, N. & Gafner, J. (2004) Identification and analysis of 4-ethyl
catechol in wines tainted by Brettanomyces off-flavor. American Journal of Enology and
Viticulture. 55, 304A.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Renouf, V., Louvaud-Funel, A. & Coulon, J. (2007) The origin of Brettanomyces bruxellensis in
wines: A review. Journal Internationale de Science de Vigne et Vin. 41, 161 - 173.
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of "Brett character". Food Chemistry. 114, 15 – 19
.
148
Chapter 7: 8BGeneral discussion and conclusions
This study aimed to investigate in detail the sensory effects of several Brett-related spoilage
compounds by first identifying their detection thresholds, profiling them separately and profiling
them in combination. All sensory experiments in this study were performed on Pinotage red
wine spiked with the relevant compounds.
When establishing the detection thresholds of eight compounds (4-ethylphenol, 4-
ethylguaiacol, 4-ethylcatechol, isovaleric acid, isobutyric acid, 4-vinylphenol, 4-vinylguaiacol and
acetic acid) it was found that large variation existed in the ability of judges to detect these
different compounds. This is in agreement with the findings of Curtin et al. (2008) regarding
Brett-related compounds. It was also found that using the median for these determinations was
a better way of dealing with this type of data than traditional mean-based methods, as the
median gave a better indication of the distribution of detection throughout the population.
The differences between the detection thresholds found in this study and those in
literature could generally be explained by the difference in medium, which underlines the
importance of the determination of detection thresholds in the relevant medium before
undertaking a major sensory study. Two major discrepancies were however found considering
the literature values traditionally quoted. These were 4-ethylphenol (195 vs 605 ug/L found by
Chatonnet et al. (1992)) and 4-ethylcatechol (385 vs 60 ug/L found by Hesford and Schneider
(2004)). However, both values found in this study fell closer to values more recently determined,
namely 368 µg/L for 4-ethylphenol (Curtin et al., 2008) and 100 – 400 µg/L for 4-ethylcatechol
(Larcher et al., 2008).
Only four compounds – 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric
acid – were investigated in the remainder of the study. This choice was made as these four
compounds are most commonly linked to Brett character. Limiting the number of compounds to
be profiled in combination also produced a sample-set that could be easily managed by the
sensory panel.
Profiling 4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol and isovaleric acid produced
predictable results. An increase of 4-ethylphenol concentration caused an increase in
Elastoplast™ and leather-like attributes and an increase in 4-ethylguaiacol concentration
caused an increase in medicinal and smoky attributes. An increase in 4-ethylcatechol
concentration caused an increase in the savoury attribute, and an increase in isovaleric acid
concentration cause an increase in the pungent attribute. All four these compounds were found
to suppress berry-like character in the wines, and to produce a sick-sweet, confected aroma.
This aroma was concluded to be the direct result of the suppression of the natural berry-like
character in wines.
However, when these compounds were profiled in combination, their combined effect
was significantly different than their effect when present individually, and several second- and
149
third-order interactions were found. The most notable of these were the third-order interactions
of 4-ethylphenol*4-ethylguaiacol*4-ethylcatechol and 4-ethylphenol*4-ethylguaiacol*isovaleric
acid. It was found that both 4-ethylcatechol and isovaleric acid enhanced the Elastoplast™
attribute when 4-ethylguaiacol was present at its detection threshold. 4-ethylguaiacol was also
found to enhance the Elastoplast™ effect when 4-ethylcatechol and isovaleric acid were above
their detection thresholds. Furthermore, 4-ethylphenol, 4-ethylguaiacol and 4-ethylcatechol had
different effects on the overall aroma of the wine below detection threshold than when above
detection threshold. This is in line with the findings of Anatosova et al. (2004). We further found
that 4-ethylphenol enhanced the pungent descriptor that is associated with isovaleric acid.
These three interactions could be partially responsible for current unanswered questions and
the controversy questions surrounding sensory Brett character, and therefore require further
investigation.
Consumer analysis of wines with Brett character was largely inconclusive. It was found
that the sample with the highest level of 4-ethylphenol was least liked by consumers, while the
“control” sample was liked less by consumers than some of the samples with a degree of Brett
character. It can therefore be concluded that Brett character can in some cases increase
consumer liking of wine.
Although not the first study conducted on Brett character, this study is unique because it
systematically investigated four Brett-spoilage compounds, whereas previous studies either
omitted 4-ethylcatechol (Romano et al., 2009) or isovaleric acid (Curtin et al., 2008). However,
due to the interactions found in this study, it can be concluded that all four compounds have a
significant effect on Brett character. It is also likely that not only these four compounds interact
in terms of Brett character, and that other compounds are also strongly involved. It is therefore
recommended that future studies should include at least all four these compounds. This is of
particular importance in the South African wine industry, as Pinotage, a uniquely South African
cultivar, contains high levels of the precursors of 4-ethylcatechol (De Villiers et al., 2005) and
could therefore be more susceptible to spoilage by this compound.
This study was the first performed in South Africa to include the chemical analysis of 4-
ethylcatechol in red wine. Although a strong correlation was found between the levels of 4-
ethylphenol and 4-ethylguaiacol (and isovaleric acid) present in samples, no correlation was
found between the levels of these compounds and 4-ethylcatechol. The data suggests that
cultivar or strain of Brettanomyces may play a role in this lack of correlation, which is in
agreement with the results of Larcher et al. (2008). The next task would be to undertake a large-
scale investigation into the prevalence of 4-ethylcatechol in South African red wines, especially
Pinotage.
Although the present extensive investigation of the sensory effects of Brett character has
shed some light on the complexity of the phenomenon, several important unanswered questions
remain. While it has been known for many years that Brett character is complex, we now have a
150
better understanding regarding the reason for this complexity. This study thus paved the way for
future research of the sensory effect of Brett infection in wine.
It is recommended that follow-up studies should be done on the same subject matter.
This should take the form of a sensory investigation of the effects of these four compounds in
the form of a factorial design (all combinations of all compounds to be tested). However, such a
study would be limited in its concentration range, as the number of levels to be tested increases
the number of samples exponentially. For instance, if only 3 levels were to be investigated, 81
samples would have to be studied, which is already a vast number for descriptive sensory
analysis.
It is generally suggested that all future studies – whether sensory, chemical or
microbiological – regarding Brett character should include analysis of 4-ethylcatechol and
isovaleric acid. It has been shown in this study that these compounds do interact with the other
spoilage compounds in terms of sensory effects. Any studies failing to include these compounds
therefore potentially lack important information.
REFERENCES
Chatonnet, P., Dubourdieu, D., Boidron, J. & Pons, M. (1992) The origin of ethylphenols in
wines. Journal of the Science of Food and Agriculture. 60, 165 - 175.
Curtin, C. Bramley, B. Cowey, G. Holdstock, M. Kennedy, E. Lattey, K. Coulter, A. Henschke, P.
Francis, L. Godden, P. Sensory perceptions of 'Brett' and relationship to consumer
preference. Blair, R.J.; Williams, P.J.; Pretorius, I.S. Proceedings of the thirteenth
Australian wine industry technical conference, 29 July-2 August 2007, Adelaide, SA. :
207-211; 2008.
De Villiers, A., Majek, P., Lynen, F., Crouch, A., Lauer, H. & Sandra, P. (2005) Classification of
South African red and white wines according to grape variety based on the non-coloured
phenolic content. European Food Research and Technology. 221, 520 - 528.
Hesford, F. & Schneider, K. (2004) Discovery of a third ethylphenol contributing to
Brettanomyces taint. Obst- und Weinbau. 140, 11-13.
Larcher, R., Nicolini, G., Bertoldi, D. & Nardin, T. (2008) Determination of 4-ethylcatechol in
wine by high-performance liquid chromatography-coulometric electrochemical array
detection. Analytica Chimica Acta. 609, 235 - 240.
Romano, A., Perello, M. C., Lonvaud-Funel, A., Silcard, G. & de Revel, G. (2009) Sensory and
analytical re-evaluation of “Brett character”. Food Chemistry. 114, 15 - 19.